feat(presenter): isolate audio delay controls for redesign

[Shinokawa]

Summary

Extracts the audio-delay portion of #53 into its own draft so the API wiring and clock design can be reviewed without blocking mute or subtitle work.

The current implementation carries the original approach: it shifts decoded audio-frame PTS before the output ring buffer and wires setAudioDelay through the full platform stack.

Draft blockers

This implementation is intentionally not ready to merge:

• samples remain in FIFO order; retagging PTS shifts the audio-master clock instead of delaying or advancing sample output
• negative offsets beyond the 250 ms stale-clock tolerance are rejected
• startup, seek, EOF, reported-position, and queued-audio transition semantics are not strict mpv-style audio delay

The intended redesign is an audio delay processor with separate source and presentation timelines: positive changes insert output silence, negative changes trim queued samples, and the master position remains monotonic and truthful.

Validation

• cargo fmt --all -- --check
• cargo test -p erika -p erika_capi --lib — 391 + 26 tests passed
• flutter test — 37 tests passed

Split from #53; draft status documents the known semantic blocker.