feat(moq-ffi)!: defer dynamic track accept so media tracks declare a timescale

[kixelated]

Summary

A track handed out by MoqBroadcastDynamic.requested_track() was accepted eagerly with no timescale. That's right for raw tracks (moq-boy status/command), but wrong for media: a media track should declare the container timescale (hang::container::TIMESCALE) so per-frame timestamps ride the moq-net wire and the relay gets timing without parsing the payload.

The constraint is that the accept freezes the timescale into TrackInfo and a subscribe blocks until that accept, so the timescale must be chosen at the accept and can't be amended afterward. One generic requested_track() can't satisfy both raw (untimed) and media (timed) with a single hardcoded accept.

This PR defers the accept to the first producer operation, which knows the right timescale:

• requested_track() returns a not-yet-accepted handle (wraps a TrackRequest).
• publish_media_on_track accepts at the media timescale → the requested media track is now properly timed.
• raw write_frame / append_group / consume / used / unused / finish accept untimed.
• abort rejects a still-pending request (the subscriber's subscribe fails), or aborts an already-accepted one.

This is the follow-up to #1750's quick fix (which only made media tolerate an untimed requested track); the track is now timed at the source, as the maintainer's timestamp-on-wire convention wants.

Breaking change (behavior, not signature)

No FFI signatures change. The observable difference: the requesting subscriber now stays pending until the first producer op on the requested track, instead of unblocking at requested_track(). This is why it targets dev. swift/kt/go auto-bindings and libmoq are unaffected (no signature change; verified libmoq has no refs and builds clean). The kt/swift dynamic-broadcast doc examples (writeFrame / abort in a requestedTracks loop) remain valid.

Python

• Broadened subscribe_media to also accept a bare Container — the consumer-side counterpart of the dynamic flow, where you subscribe before the catalog exists.
• Restructured the dynamic tests to run the subscribe concurrently. This also fixes test_dynamic_track_request_can_publish_media, which never actually ran before (it was missing awaits and passed a Container where the wrapper wanted a catalog record).

Test plan

• Rust: 22 moq-ffi tests pass (incl. all three dynamic_track_request*); clippy + fmt clean via nix.
• Python: FFI wheel built via maturin; 31 local tests pass (incl. both dynamic tests); ruff + pyright clean.
• libmoq + moq-ffi build clean; rebased onto current dev (post per-frame-duration removal) and re-verified.

🤖 Generated with Claude Code

(Written by Claude)