refactor: carry moq-video decode timestamps as moq_net::Timestamp

[kixelated]

Summary

The moq-video decode path carried presentation timestamps as a bare u64 of microseconds, the one spot in the media pipeline not already using moq_net::Timestamp — the container frame is container::Frame.timestamp: Timestamp and encode::Producer::publish takes a Timestamp too. This threads Timestamp through decode so it is symmetric with encode and the container.

Note: this PR was originally about surfacing per-output decoder PTS through the backends. That work landed independently on dev via #2145 (which added the NVDEC backend and the zero-copy NVDEC → NVENC path). Rebased onto #2145, the only remaining, non-redundant improvement is the timestamp typing, so the PR is now scoped to that.

What changes

u64 microseconds → moq_net::Timestamp across the decode surface:

• decode::Frame.timestamp, decode::Decoder::decode, and the pub(crate) Backend::decode / Decoded.
• All four backends: openh264 / VideoToolbox / Media Foundation echo the input timestamp (unchanged behaviour, just retyped); NVDEC continues to thread it through the cuvid parser (Timestamp::as_micros() in, Timestamp::from_micros() out).

Deletes three lossy conversions that only existed to bridge the u64 gap:

• decode::Consumer::read now passes mux_frame.timestamp straight through (was as_micros().try_into().map_err(TimeOverflow)).
• moq-transcode's rung serving drops an as_micros() on the way in and a from_micros() on the way out; the decode→encode timestamps now flow as Timestamp end to end.

libmoq's C ABI is unchanged (moq_video_frame.timestamp_us: u64); the adapter converts via frame.timestamp.as_micros().

Public API change (moq-video)

• decode::Frame.timestamp_us: u64 → decode::Frame.timestamp: moq_net::Timestamp
• decode::Decoder::decode(payload, timestamp_us: u64, keyframe) → decode(payload, timestamp: moq_net::Timestamp, keyframe)

Breaking reshapes of pub items, but moq-video/moq-transcode are unreleased (dev-only, 0.0.x) with no published contract, so this lands on dev. No wire-protocol or C-ABI change.

Test plan

• macOS: cargo build/test -p moq-video; decode round-trip tests pass (VideoToolbox H.264 + H.265, openh264), now also asserting the timestamp round-trips through the codec and returns strictly increasing.
• macOS: moq-transcode builds; clippy clean (moq-video --all-targets, moq-transcode --lib); rustdoc -D warnings clean; cargo fmt applied.
• Windows: MediaFoundation backend cross-compiles via cargo zigbuild --target x86_64-pc-windows-gnu.
• Not verified locally (mechanical edits, left to CI): the NVDEC backend (Linux + CUDA) and libmoq — both are un-buildable on macOS on this branch independent of this change (a pre-existing dev feature/target quirk), so CI is the check for them.

🤖 Generated with Claude Code

(Written by Claude Opus 4.8)

[sourcery-ai]

Sorry @kixelated, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[coderabbitai]

Caution

Review failed

An error occurred during the review process. Please try again later.

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch claude/moq-video-deferred-work-a32311

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}