Saner defaults for avvidec threading
Since !34 (merged) was merged, the default threading models of decoders is a bit painful.
If upstream is not live, it defaults to:
- up to 16 threads
- frame and slice threading
What actually happens is that most of the time ffmpeg will end up picking only frame-based threading and backlog up to 15 frames before outputting anything, thereby introducing a massive latency.
Furthermore, we already have frame-based threading upstream and downstream of decoders.
What I propose is that we change the defaults to:
- Maximum of 4 threads if
max-threads
is 0 - Use slice-based threading (if available) if
thread-type
is 0 (auto)