perf(studio): reuse the audio mix across optimistic edits

[LeadcodeDev]

Severity Low, category quality. Location: crates/rustmotion-studio/src/editor/playback.rs:115

Impact

optimistic::commit bumps generation on EVERY mutation (one per oninput event), and use_hot_reload polls it every 250 ms, so a slider drag on a scenario that has audio calls audio::prepare ~4x/second. prepare spawns a brand-new std::thread that runs mix_audio_tracks over the whole scenario — documented in audio.rs:158 as "seconds on a long scenario" — decoding and resampling every track from scratch even though a font-size edit cannot have changed the audio. MIX_TOKEN only discards the RESULT; nothing cancels the work. Dozens of concurrent decode threads pile up, each holding a full interleaved f32 PCM buffer (~23 MB for 60 s stereo), fighting the 2-6 prefetch render workers for cores during exactly the interaction that most needs to stay smooth.

Fix

Only call prepare when the audio actually changed — compare a hash/fingerprint of scenario.audio plus total_duration against the last prepared one and return early otherwise. Optionally serialize mixing onto a single dedicated worker so a burst cannot spawn N threads.

Evidence the audit read

if last_gen != Some(g) {
    if last_gen.is_some() {
        rev.set(rev() + 1);
    }
    last_gen = Some(g);
    let fps = scenario.video.fps.max(1);
    super::audio::prepare(scenario, total as f64 / fps as f64);
}

Stacked on fix/studio-render-panic, which carries the previous finding of this workstream. GitHub shows only this finding's diff; merge in order.

Part of the September 2026 audit remediation chantier. Refs #220 (RM-39).