fix(audio): eleven small correctness fixes across engine, core and the panel

[vanceingalls]

First of four slices of the old #3156, cut to stay under 1000 lines.

Small, independent fixes with no audible change:

• an empty audio track is handed back instead of failing the mix
• the browser lease is taken inside the try that releases it
• the phaser's trim lanes aim at the trims, not the pinned wet/dry pair — an envelope was modulating a constant while the trim it was meant to move stayed frozen
• a node's id moves with its slot on an in-place chain update, so a lane cannot end up addressing its neighbour
• the volume lane is sampled at clip-local time, the way the render does it
• an FX parameter can be typed, and a bare blur no longer re-persists
• a clip's FX graph is disposed when it ends, not only when the element goes
• the rack locks while the carve measures

Each commit body carries its falsification note — the test was broken on purpose and confirmed to fail.

🤖 Generated with Claude Code

[miga-heygen]

Review: fix(audio): eleven small correctness fixes — #3173

Verdict: LGTM

Eleven fixes, each independently falsified — the commit notes prove each test broke when the old behavior was restored. Highlights:

The signal routing fix is the biggest silent one. effectiveSignal now correctly propagates to downloadToTemp, the video extract, and the trim. The old code passed the caller's signal to all three, so internalController.abort() cancelled nothing — and the rmSync(workDir) on the next line ran while their ffmpeg children were still writing into it. This is a genuine race that could corrupt renders under cancellation.

Browser lease ordering — moving acquireBrowser inside the try that releases it is correct. A mkdtempSync failure between the old acquire and the try leaked a pooled browser, and a pool with no leases left hangs every subsequent render.

Phaser trim targets — asserting by VALUE rather than node identity is the right call. The trims are internal; the honest question is whether the param the lane drives is the one the knob sets. Clean.

FxParamRow typing fix — the typing state as TEXT (not number) is exactly right: a half-typed "-" or "" is a valid intermediate state, not a value to clamp. The edited ref preventing no-op commits is a nice touch — prevents a full source-patch/selection-resync/preview-reload/audio-restart cycle for a gesture that moved nothing.

Lint rule regex — [^;)] instead of [^;] to stop the match from crossing ) into a later .to() call in a chained timeline. The false-negative direction (missing a case where a call result contains )) is the safe one for a rule that guesses.

No issues found.

---

Review by Miga

🤖 Generated with Claude Code