feat(pvlint): PV009 — gain MULTIPLIED by a control that seeds to zero, i.e. a layer that never arrives

PLN, mid-run, working it out himself: "d9 on wap didnt take effects on 17 and 19?
ok no its the 17 superimpose that melts too !! does our xfade screw the
superimposes? 💡"

The xfade is innocent. wap.tidal:91 is

    $ superimpose ( (# "acidOto3091") . (|- note 12) . (|* gain "^17") . (# cut 91) )

and ^17 seeds to 0 under BootTidal's policy ("A/B knobs -> 0, effect amount off").
Gain MULTIPLIED by 0 is silence, so the superimposed layer is not quiet — it does
not exist. The 4-cycle dN crossfade then eases from the previously-audible state
into that silence, which is what reads as a melt and is exactly why the xfade
looked guilty. It was the messenger.

The distinction the rule encodes: `#` ASSIGNS and survives a 0 seed
(`# crushbus 51 (range 16 4.5 "^53")` at seed 0 is simply crush 0), while `|*`
and `|/` ANNIHILATE. Only the multiplicative form can delete a pattern, which is
why flagging `#` sites would bury the real finding under hundreds of false ones.

SEVERITY IS `info`, DELIBERATELY. This idiom is usually correct — it is a bring-in
fader for a layer, and PLN uses it that way. The rule does not exist to call it
wrong; it exists so nobody hunts it BY EAR. Making it a warning would fire on a
normal, intended idiom and get the whole gate ignored, the mistake already made
once in check-drift's first draft. It shows under `pvlint --info`.

WHY IT MATTERS MOST RIGHT NOW: a remap moves the CC, but it does not move the
KNOB. Physical positions survive across a surface change, so a knob left at 0
silently acquires a new victim. That is the single root cause behind everything
PLN heard this session — d9's layer gone on wap, and the same class of staleness
behind the d4/d5/d8 symptoms (latched buttons whose orbit changed under them). It
is the identical failure to the very first question he asked today: Ardour's fader
raised by mouse while the physical fader sat at 0. There are no motors, so the
surface never follows the map.

SCOPE, measured (both forms — `|* p (range 0 hi "^NN")` and the bare
`|* p "^NN"`, which my first scan missed and which is the worse of the two):
    33 sites across 25 files corpus-wide
     2 in the OPAL setlist:  wap.tidal:91 (bare), desire.tidal:75 (range 0 1.5)
Notable elsewhere: scratchomatic multiplies gain by "^84" — an ARDOUR-learned
fader — and computer_riddim/computer_riddub/anniv multiply by "^50", the gF2 DJ
filter knob.

6 new tests. The negatives carry the weight, as always: `range 1 1.5` (the
prescribed neutral-low-end fix) must NOT flag, `#` assignment must not flag, and
C1-C3 must not flag because they seed to 0.5 (centre), not 0. 496 pass.