fix(player): drive playback through the audio engine, not the silent multitrack

[thcp]

Fixes #515. Independent; branches off 0.16.1.

The bug

engineMode() returns "chunked" unless the user set localStorage["stemdeck.audioEngine"] = "0". On that path audioEngine owns the clock and the multitrack is mounted with url: null for visuals only -- its own comment says so: "The engine emits no play/pause events (the multitrack stays silent)".

transport.js handled this everywhere via audioEngine ?? multitrack (lines 74, 386, 423, 596). main.js imported only multitrack and called it bare.

On the default configuration, therefore:

control	behaviour
Footer scrub bar	nothing at all -- full-size cursor: pointer overlay, click or drag moved neither playhead nor audio
[ / ] (seek +/-5s)	nothing
I (loop in at playhead)	always wrote 0, regardless of playhead position
O (loop out at playhead)	always wrote max(0, loopStart + 0.5)

Space and ruler clicks were fine -- they live in transport.js.

The fix

Rather than patching five call sites, transport.js now exports the accessor it was already using internally, plus setPlayheadTime.

Seeking goes through setPlayheadTime, which also updates the playhead marker, footer times and presence playhead. The old multitrack.setTime call did none of that, so the scrub bar would have left the marker stale even if it had worked.

Same anti-drift shape as apply_ffmpeg_path in #506: one accessor, no re-deriving.

multitrack is no longer imported by main.js at all.

Also fixed

The keydown guard excluded HTMLInputElement but not HTMLTextAreaElement, so Space in the settings log viewer started playback instead of scrolling. Filed as a separate low finding in #515; fixed here since the guard was being edited anyway.

Verification

New tests/js/transport-clock.test.mjs. It is structural rather than DOM-driven on purpose: the bug was not a function returning a wrong value, it was reaching for the wrong object. It asserts main.js contains no bare multitrack.setTime / getCurrentTime / getDuration, that the accessor exists and prefers the engine, and that the textarea guard is present.

Confirmed not vacuous -- reverting main.js and transport.js fails all 8 checks.

node tests/js/transport-clock.test.mjs   8/8
npm run test:js                          48/48
node --check                             main.js, transport.js OK

One thing I did not do

There are 83 Playwright e2e tests but none covers the scrub bar. A real drag-and-assert-playhead test would be stronger than the structural check. I did not add one because the e2e suite needs a processed track fixture and that is a bigger lift than this fix warrants -- worth knowing the gap exists.