This change enables feeding decoders from the closest sync frame
before a specified seek position, where-as previously we'd
always feed decoders from the start of the chunk. This avoids
decoding and discarding many audio samples during each seek. The
same benefit also applies to video chunks containing more than
one key-frame.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=190539547