First slice of the #92 column migration, and the slice that had live breakage in it.

THE GRID, settled with PLN this afternoon after one wrong turn and one measurement
that overruled me:

    A 13-16  d9 d10 d11 d12 LEVEL    (Ardour, learned today)
    A 17-20  d9 d10 d11 d12 EFFECT   <- this commit
    B 29-36  d1..d8 EFFECT           (next slice)
    C 49-51  gF1 gF2 gF3             (unchanged — see below)
    C 52-56  free, per-orbit 2nd effects
    D 77-84  d1..d8 LEVEL            (Ardour, learned today)

The wrong turn was mine: I proposed extending the DJ filters to gF1..gF8, one per
column. Then I measured which orbits actually receive each filter across all 13
setlist tracks:

    gF1 -> d1×13  d2×13  d3×13  d8×12     drums / core rhythm
    gF2 -> d4×14                          bass, almost exclusively
    gF3 -> d5×12  d7×9  + d9-d12          leads / melodic / extras

That is not three arbitrary filters, it is a per-FAMILY design applied consistently
across the whole set — you filter a stem group, like a DJ mixer, not one channel.
Per-orbit filters would have destroyed it and handed PLN eight knobs he would never
use that way. gF1-3 stay exactly as they are, and C4-C8 stay free, which as a bonus
removes the "one effect per orbit" constraint I thought the grid forced. PLN's
instinct ("i wanna keep C1/2/3 as djf these are core") was righter than his
reasoning for it.

== THE LIVE BREAKAGE ==
piment_bresilien.tidal:77 drove its d10 crushbus from "^14". As of today CC 14 is
MIDI-learned to Ardour's Tidal 10 gain. So that one knob was crushing the bass and
riding the orbit's fader simultaneously — a control doing two unrelated jobs, the
kind of thing that reads as "the rig is haunted" mid-set. Now ^18 (A6), which is
d10's own effect slot. This was the concrete half of #88.

== THE REST ==
  wap        d9  ^18 -> ^17     gain swell
  bombe_dj   d9  ^18 -> ^17     midiOff (slow 4)
  desire     d9  ^19 -> ^17     gain swell
  do_it_right d12 ^31 -> ^20    modIndex  (^31 was B3 = d3's slot, and was also
                                Ardour-bound until this afternoon's release)

== ONE BEHAVIOUR CHANGE THAT NEEDS PLN'S EAR ==
In desire.tidal, "^19" drove BOTH d9's gain swell AND d7's NANANA (line 84). Moving
d9 to its own knob necessarily splits that gesture. If the link was deliberate —
one knob swelling the synth while bringing in the vocal — it is now two knobs and
we should restore it. Flagged with a FIXME in the file rather than silently
"fixed", because a lost gesture is not something a linter can miss for you.
d7's ^19 moved to ^55 (C7, its own 2nd-effect slot) rather than squatting on A7,
which belongs to d11.

== FIXMEs LEFT ON PURPOSE (PLN: "if we have two effects, we consolidate via gSel
   indeed. mark FIXMEs in code in there") ==
wap d9 and bombe_dj d9 each have a SECOND effect (crushbus, modIndex) and d9-d12
get one A-row knob each, so there is no honest home for it. Both marked
FIXME(#54) to fold into ^17 via gSel rather than borrowing d11's knob. The wrong
fix would have been to spread them onto neighbouring columns and call it done.

== VERIFICATION, in the order that matters ==
  pvlint            5 tracks, 0 errors (2 pre-existing warnings, untouched)
  silent-eval       every declared orbit of all 5 tracks still emits events, cold,
                    against the seeded control map — this is the one that proves a
                    renumber did not silently empty an orbit
  surface-columns   Ardour collisions 1 -> 0; the 5 primary d9-d12 controls all
                    land on their own knob; 2 second-effects correctly flagged

Uncommitted WIP in 6 setlist tracks was backed up before editing; only CC numbers
and comments changed, so PLN's in-progress edits are preserved in the diff.

Aside, spotted by pvlint and NOT fixed here: do_it_right d5 references index 8 of
"daft" which has 5 samples — that layer never arrives. Own task.