fix(export): WebCodecs video decode and full audio mixdown

[dazzatronus]

What

Rebuilds the SDK's client-side video export, which produced broken output: video froze on its first frame and stuttered, audio was truncated to ~2% of its length and mis-placed, and audio embedded in video clips was dropped (only standalone audio clips were encoded).

Changes

• Audio — replace the per-clip AudioSampleSource (which can't mix) with an OfflineAudioContext mixdown. Every audio-bearing clip (audio + video + text-to-speech) is scheduled with per-clip gain automation (volume + fadeIn/fadeOut) and rendered to one AudioBuffer, encoded via AudioBufferSource. Fixes the seconds/ms timing bugs (length, placement, trim) and an AudioSample leak; overlap is summed, gaps are silence, differing sample rates resample natively.
• Video — decode source frames with mediabunny Input + CanvasSink (WebCodecs) instead of seeking an HTMLVideoElement. Frame-accurate, no stutter, and decodes in browsers that can encode but can't <video>-play the source codec. Adds a canDecode() capability gate; removes the per-frame texture churn and a GPU texture leak.
• Tests — pure-function unit tests for the timing and volume-automation math.

Testing

• tsc, eslint, and jest (1801) pass.
• e2e: exported an edit covering 15 asset types (validated against the Edit API with shotstack validate); all types export, audio mixes correctly (video soundtrack + music), and source frames decode 125/125.

[github-actions]

🎉 This PR is included in version 2.11.4 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀