SDK / exoplayer

Currently, MetadataRetrieverImpl uses MediaCodecVideoRenderer to renderering
frames to capture. It suffers from 2 issues:
- MediaCodecVideoRenderer has a lot of logic to time frame release, including
drop frames if late or try to keep up with feeding speed etc... These are not
necessary at all for frame retrieval.
- MediaCodecVideoRenderer always tries to render 1 frame after video surface
update. This may some times lead to the wrong image being rendered for
MetadataRetriever.
This CL replaces this with a custom video renderer, that does not time frame
releases, and only render first frame after position reset / renderer
re-enabling.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=207276937