SDK / exoplayer

- Remove skipping of the VBRI/XING frame before calculating position
  offsets. This was incorrect. Instead, a constraint is used to ensure
  we don't return positions within these frames, the difference being
  that the constraint adjusts only positions that would fall within
  the frames, where-as the previous approach shifted positions through
  the whole stream.
- Excluded last entry in the VBRI table because it has an invalid
  position (the length of the stream).
- Give variables in XingSeeker descriptive names.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177451295

The set of active audio processors was only updated on reconfiguration and when
draining playback parameters completed. Draining playback parameters are cleared
in reset(), so if parameters were set while paused then the sink was quickly
reset, without draining completing, the set of active audio processors wouldn't
be updated. This means that a switch to or from speed or pitch = 1 would not be
handled correctly if made while paused and followed by a seek.

Move resetting active audio processors from configure (where if the active audio
processors were reset we'd always initialize a new AudioTrack) to initialize().

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177442098

- This change snaps the seek position for constant bitrate MP3s
  to the nearest frame boundary, avoiding the need to skip one
  byte at a time to re-synchronize (this may still happen if the
  MP3 does not really have fixed size frames).
- Tweaked both ConstantBitrateSeeker and WavHeader to ensure the
  returned positions are valid.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177441798

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177430827

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177424314

Support ad MediaSources that aren't prepared immediately by using
DeferredMediaPeriod, moved up from DynamicConcatenatingMediaSource.

In a later change the new interfaces will be made public so that apps
can provide their own MediaSource factories.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177424172

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177419981

captions fetcher architecture.

1.  ManifestlessCaptionsMetadata

Other captions fetchers must first fetch a manifest (HLS or manifest) to
discover captions tracks.  This process does not exist for manifestless.  All
we need to do is scan the FormatStream's for the right itag, so this is an
all-static class.

2.  ManifestlessSubtitleWindowProvider

Once a captions track is selected, a subtitles provider is instantiated.  This
is the main interface used by the player to retrieve captions according to
playback position.  This class stores fetched captions in a tree index by time
for efficient lookups.  Background captions fetches are used to populate
the tree.

3.  ManifestlessCaptionsFetch

Captions are fetched one segment at a time.  One instance of this object
is required per fetch.  It performs a blocking fetch on call(), and is
intended to be submitted to a background-thread executor.

4.  ManifestlessCaptionsFetch.CaptionSegment

This is the result of the caption fetch.  These values are used to populate
the captions tree.

Manifestlessness

The initial request is always a headm request.  There is a separate tree
of every segment indexed by start time.  This tree is used to improve
manifestless sequence number calculation.  Once we have data for the current
timestamp, we walk forward through the tree to find the next unfetched
sequence number, and fetch that.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177385094

- Avoid re-downloading data prior to the first mdat box when
  seeking back to the start of an unseekable FMP4.
- Avoid re-downloading data prior to the first frame for
  constant bitrate MP3.
- Update SeekMap.getPosition documentation to allow a non-zero
  position for the unseekable case. Note that XingSeeker was
  already returning a non-zero position if unseekable.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177317256

There are still things broken about the seeker, but this
cleans up some of the weird bits.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177315136

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177184331

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177175377

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177173618

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177170994

- Align class summary Javadoc
- Fix ErrorProne + Style warnings

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177165593

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177141094

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177063576

Merge of https://github.com/google/ExoPlayer/pull/3495

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177057183

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177011497

Also slightly improve language normalization/documentation.

For this CL, it is assumed that null and "und" languages are different
entities. Once we fully tackle language tag normalization, we can decide
whether to normalize the "undefined" language.

Issue:#2867
Issue:#2980

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=177008509

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176997767

A message to stop the playback and to quit the playback thread was posted in release().
The stop message removed all other already queued messages which might include
the second message to quit the thread. That led to infinite waiting in the
release method because the playback thread never got the quit signal.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176997104

currentTimeMillis is not guaranteed to be monotonic and elapsedRealtime is
recommend for interval timing.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176853118

In a test run where no exceptions were thrown on the main thread and the test
did not time out, exceptions from onPlayerError were not correctly propagated to
the test thread (handleException would be called with null).

Fix ExoPlayerTestRunner.onPlayerError to propagate the actual exception from the
player.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176825907

We've found that in our production environment, the AAC stream's timestamp exceeds the 33bit limit from time to time, when it happens, `peekId3PrivTimestamp` returns a value that is greater than `TimestampAdjuster.MAX_PTS_PLUS_ONE`, which causes a overflow in `TimestampAdjuster.adjustTsTimestamp` (overflow inside `ptsToUs`) after playing for a while . When the overflow happens, the start time of the stream becomes negative and the playback simply stucks at buffering forever.

I fully understand that the 33bit is a spec requirement, thus I asked our stream provider to correct this mistake. But in the mean time, I'd also like ExoPlayer to handle this situation more error tolerance, as in other platforms (iOS, browsers) we see more tolerance behavior.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176693785

Parse DASH manifest's publishTime node as defined by ISO/IEC 23009-1:2014,
section 5.3.1.2.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176525922

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176366693

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176366471

Issue:#3478

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176351086

Issue: #2465

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176341309

In particular this allows to have the workaround for
https://code.google.com/p/dexmaker/issues/detail?id=2 in one place only.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176340526

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176335667

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176333544

Add Builder pattern to SingleSampleMediaSource and mark existing constructors as
deprecated.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176332964

- MediaSourceTestRunner aims to encapsulate some of the logic
  currently used in DynamicConcatenatingMediaSourceTest, so it
  can be re-used for testing other MediaSource implementations.
- The change also fixes DynamicConcatenatingMediaSourceTest to
  execute calls on the correct threads, and to release handler
  threads at the end of each test.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176117535

This test seems to obtain a timeline from a prepared
FakeMediaSource, but that's identical to the timeline
passed into that source to start with :).

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=176117133