fix(producer): bound nested media by a host's data-duration when it has no data-end

[valeriangalliat]

What

In the render media planner, a composition host authored with data-duration but no data-end now closes at data-start + data-duration, so the audio and video nested inside it stop at the slot end.

Fixes #3964.

Why

resolveHostWindow bounded a host by data-end only, and compileTimingAttrs never stamps data-end on composition hosts, so a slot authored with data-duration was unbounded for the planner: a scene's audio kept playing over the next scene while the runtime (since #1662) already hid the slot's timed descendants at start + duration. Pixels and audio disagreed about where the slot ends; preview and render disagreed too, since the runtime media sync honours the authored host window.

How

renderMediaCollector.ts gains resolveHostEnd(host, hostStart): data-end wins, else hostStart + data-duration, else unbounded as before. hostStart is the already-resolved start, so id-ref hosts (data-start="intro") are bounded correctly too. Open-ended media inside a bounded host now closes with the host instead of staying end: 0.

One existing compile test (offsets template-wrapped media to the host start during compile) asserted the unbounded end — a 4s clip in a 2s slot ending at 6 — and now expects the slot end (4).

Test plan

• Unit tests added/updated: new closes a host authored with data-duration but no data-end case (numeric and id-ref hosts, a clip starting after the slot is dropped); legacy-global and template-wrapped tests updated for the bounded end.
• Manual testing performed: the reproduction at https://github.com/ArcadeHQ/hyperframes-repros/tree/patch/host-duration-bound renders −21 dB over the following scene on 0.8.40 and −91 dB (silence) with this change; frames unchanged.
• Documentation updated (not applicable)