Make mp4 parsing **a lot** faster & tremendously lower memory overhead

[jprochazk]

What

• Part of Optimize mp4 parse times, and avoid duplicating their RAM use #7481
• Depends on Optimize mp4 parse times by not copying video data re_mp4#12

Some rough numbers:

main@de4c389   297.43   µs
this branch      3.8957 µs

Measured on Big_Buck_Bunny_1080_10s_av1.mp4 from our test assets (downloaded via tests/assets/download_test_assets.py)

Total Rerun memory usage for a 297 MiB video (the entire Sintel movie):

main@de4c389   573 MiB
this branch    325 MiB

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested the web demo (if applicable):
  • Using examples from latest main build: rerun.io/viewer
  • Using full set of examples from nightly build: rerun.io/viewer
• The PR title and labels are set such as to maximize their usefulness for the next release's CHANGELOG
• If applicable, add a new check to the release checklist!
• If have noted any breaking changes to the log API in CHANGELOG.md and the migration guide

• PR Build Summary
• Recent benchmark results
• Wasm size tracking

To run all checks from main, comment on the PR with @rerun-bot full-check.

[jprochazk]

We still copy each sample when we enqueue a chunk, which feels like it should not be necessary. It would be nice to have direct access to the video data arrow buffer in decoders, so that we can index into it there, but I wasn't 100% sure how to make that work here so I left it out.

[Wumpf]

similar to the commentary on the parent PR I'm super happy about this being so low-hanging, but also like there the distinction of "what data is this" needs more love. In partiulcar VideoData is already a type and now we're passing a lot of video_data which is not related to it.
Ideally we'd use some reference counted object from the guts of Blob, but I haven't looked into whether that's viable (we really don't want to end up passing around raw arrow buffers, infecting everything with that dependency)

If we advertise this in the changelog, we should have before after perf & memory numbers. Just give it a quick spin with one of the larger videos on our internal video repo, thank you! :)

[jprochazk]

(trying to measure this by adding a benchmark)

[jprochazk]

Some rough numbers:

main@de4c389   297.43   µs
this branch      3.8957 µs

Measured on Big_Buck_Bunny_1080_10s_av1.mp4 from our test assets (downloaded via tests/assets/download_test_assets.py)

Total Rerun memory usage for a 297 MiB video (the entire Sintel movie):

main@de4c389   573 MiB
this branch    325 MiB

[jprochazk]

Running a nightly build to check if the bench works: https://github.com/rerun-io/rerun/actions/runs/11461677788 (it did not)

[jprochazk]

Bench was missing nasm for the build, maybe it works now: https://github.com/rerun-io/rerun/actions/runs/11461912756/job/31892063198 - will cancel it if the benchmark succeeds - it downloaded the test assets and successfully compiled, good enough

[Wumpf]

awesome!

[Wumpf]

🤯