Transpose the whole track, or one lane at a time, on one clock - #479

The per-lane VU meters were linear in amplitude, min(1, rms * 2.5).
Loudness is logarithmic and a separated stem sits around -20 to -30 dBFS,
so normal playback lived in the bottom tenth of every bar and the top
three quarters was decoration.

Measured across four tracks of a real library, median RMS per window
through the old scale:

    vocals   5.0%    drums   1.7%    bass  5.2%    other  1.6%
    vocals  20.5%    drums   8.5%    bass 34.1%    other  2.8%
    mean median 10.4%

A -60 dBFS floor mapped onto the bar is the usual choice for a small
console meter: quiet passages still register and full scale stays
reachable without being reached constantly. The same stems now read
27.8%, 26.9%, 43.4% and 71.1%.

This forced the meters off getByteTimeDomainData. Its quantisation step,
one part in 128, is -42 dBFS, which on any dB scale with a lower floor
would light every meter to about a third of full over silence.

The scale lives in its own module because player.js cannot be imported
outside a browser, and a scale nothing can measure is how the linear
version survived this long. tests/js/vu-scale.test.mjs asserts what
fraction of the bar real measured stem levels light, and keeps the old
formula as a control so the comparison is a measurement rather than a
claim about the past. It guards the other direction too: no clipping to
full, headroom preserved, and a stem 20 dB down still reads 20 dB down.

Closes #477

Adds a semitone control to the transport bar and one to every mixer
lane. The transport control moves the whole track; a lane control moves
that instrument alone, so a harmony line can be practised against the
original melody rather than instead of it.

Drums are never transposed. A resampled snare is not the same drum in
another key, it is a different drum. That refusal is enforced in three
places because there are three ways in: the UI never offers it, the
engine ignores it if a caller asks anyway, and the worklet's unpitched
input is where drums are wired regardless. It also covers drums you
cannot see: a stem deselected at import is folded into original.wav, so
playbackStems.js rebuilds that lane from the retained raw stems and
routes the drums out of it. When a job predates those stems and cannot
be rebuilt, the whole lane stays unpitched rather than risk resampling a
kit that is mixed into it.

One shared clock, not one shifter per lane. WSOLA latency moves with the
shift, so independently clocked chains put lanes up to 144 ms apart from
each other:

    pitch  0, tempo 1       0 ms   (bypass)
    pitch +2, tempo 1     115 ms
    pitch -5, tempo 1     144 ms
    pitch  0, tempo 0.75  437 ms

So the worklet takes one input per semitone, each a pitch chain, and
sums them into a single tempo stage. A lane's key is which input it is
connected to, not a parameter. Measured with five chains live at once:
0.35 cents worst pitch error, 12 ms worst skew between buses, no drift
over a minute.

Changing a lane mid-playback neither restarts nor flushes anything.
Every bus is primed with the same amount of silence, so the lane's old
chain plays out what it had buffered while the new one takes over as it
runs dry. The handover is a short duck, not a gap.

Lane keys are absolute rather than offsets from the global control, and
the global control applies its change rather than its value. So the
number on a lane is always the key that lane is in, and a lane
deliberately put a third above the rest stays a third above the rest
when the whole track moves. A reset returns the global key and every
lane at once, which stepping the global back to zero cannot do.

Three things the measurements changed along the way:

  - An earlier revision split attacks out of the music bus and passed
    them at unity while the rest went through the pitch chain. A delayed
    copy plus an undelayed copy cannot sum back to the input, and it put
    a 0.64 full-scale step into the output at every onset against a
    source whose own largest step was 0.10. Heard as static. Removed.
  - The chain produced exactly one quantum per quantum with no slack, so
    a quantum occasionally came up short and was emitted zero-padded. A
    hard edge to zero mid-block is a click. Fixed with an output reserve
    and an input cushion banked during priming.
  - The old alignment test reported 2 ms, but only because the attack
    splitter was passing its probe through the bypass. Real alignment
    was 18 ms and one-sided, being the WSOLA sequence sawtooth. It is
    predictable, so it is now cancelled rather than tolerated.

The lane row is about 300px and was already full, so the meter column
gives up 18px to keep the fader usable, and the meters take that as
height: 5px read as a hairline rather than a meter.

Closes #245
Closes #476