Let's do it this way for now. Note there's an implicit
assumption in here that non-AV tracks consist of only
key-frames, but I think we'll not encounter any issues
in the real world as a result, we already make this
assumption in ChunkSampleStream, and actually tagging
every queue with this information explicitly is a very
painful amount of plumbing.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=161545383
3c5688de73