Fit the click track to the row instead of piling it up

[thcp]

Closes #625
Closes #626
Closes #627

The transport footer ran out of width on a laptop and dealt with it by stacking the click track options into a column. This sizes the collapse to what the row is actually short, gives the volume a shape that lines up with its neighbours, and stops the loop fields cutting off their own last digit.

The collapse is measured, not a constant (#625)

footerFit had one move for the click cluster: cap the options panel at 190px and let it wrap. That is narrower than two of those controls side by side, so the wrap was a five-row column and the footer grew to about three times its height.

It now asks for exactly the pixels the row is short, floored at half the panel's inline width so it stays a two-row wrap. Past that floor the options move into a popover, opened from a disclosure beside the on/off pill. In that state the clusters spread across the row, because the popover frees more width than the row was short and the surplus used to stand as a hole before Export Mix.

Measured against the e2e fixture, gap before Export Mix:

Window	Before	Now
1920	524px, five-row popover state	fully inline
1728	332px	20px, options inline and wrapped
1600	204px	16px, popover
1440	44px	16px, popover

One thing worth flagging for review: the panel is position: fixed, not absolute. .footer-clusters is overflow-x: auto, which computes overflow-y to auto as well, so an absolutely positioned panel gets clipped by the strip it belongs to. transport.js places it from the trigger's rect, the same shape .search-panel already uses.

The volume is a pill now (#626)

It was a 116px horizontal slider with a percentage after it, on a row of 34px pills. It is a pill beside the on/off toggle, opening a vertical fader with the level above it. Same #t-metro-vol input underneath, so prefs, the K key and the export click render are untouched.

Putting it beside the toggle rather than among the options means the level stays on the row at the widths where everything else goes behind the disclosure.

The fader uses writing-mode: vertical-rl with -webkit-appearance: slider-vertical behind it. On a browser that honours neither it stays horizontal inside the popover: smaller than intended, still working.

The loop fields fit their own value (#627)

fmtTimeMs wrote nine characters into an eight-character box, so 00:56.000 displayed as 00:56.00. They now show total seconds and drop the minutes field. parseTimecode already accepted both forms, so mm:ss.mmm still types fine.

Also in here

The standing "Groups" label went: the box beside it already carries 3+2+2 as its placeholder and explains itself in its title. "Auto (detected)" became "Auto" in all ten locales.

Testing

Full local gate. npx playwright test 171 passed, npm run test:js clean, node --check clean, ruff check and ruff format --check clean, bandit clean at medium and above, uv.lock untouched, i18n coverage clean. No Rust in the diff.

pytest has two failures, test_all_stems_zip_ogg and test_ogg_is_still_streamed. Both fail the same way on a clean checkout of main, so they are not from this branch. They look like a local ffmpeg without vorbis.

Three e2e specs moved with the behaviour rather than around it:

• waitForClickTrack opens the popover when there is one, so the click-track specs keep driving real controls instead of ones that exist but cannot be clicked.
• "collapsing keeps every click control visible" became "puts them one click away", plus a new test that a slightly short row keeps them inline.
• The grouping test asserts the box's placeholder and title now that the label is gone.