96 substitutions across all 13 OPAL tracks. "Column N is orbit N" is now true for
the WHOLE surface, not just the knobs:

    BT 41-44/57-60   d1..d8 gate
    BL 73-76/89-92   gMute1/2/3 | d4..d8 gate2
    A  13-20         d9..d12 level | d9..d12 fx
    B  29-36         d1..d8 fx
    C  49-56         gF1/2/3 | d4..d8 fx2
    D  77-84         d1..d8 level

    surface-columns:  23/84 aligned, 127 renumbers  ->  82/82 aligned, 0 renumbers

PLN drove the key decision. My first attempt left over-budget controls in place and
reached only 11/13 tracks, which he rejected outright: "we need finish the job. why
half moved? cant you move all-but then comment out with fixme?" He was right on both
counts, and leaving them was not the conservative choice — it was a bug generator.

THE CASCADE, AND WHY COMMENTING OUT DISSOLVES IT
The board has 8 BT + 5 BL = 13 per-orbit buttons for 8 orbits, because BL1-3 are the
family mutes. So d1..d3 get exactly ONE button each, and bombe_dj's kick — a 3-state
gesture on TWO buttons —

    $ midiOn  ("^41"-"^42") (<| "k(3,8) ~")
    $ midiOff ("^42"+"^41") (<| "k*<1!7 2> ~ ...")
    $ midiOn  "^42"         (<| "k k k*<1!8 2!8> ...")

is over budget by design. Left in place, d1's ^42 blocked d2, which blocked d3, then
d4, then d5: 4/8 aligned. Commented out, the control is free and the row aligns.
Nothing is lost — the original sits one line above under two dashes, restorable the
moment gSel (#54) can fold three states onto d1's own control.

THREE BUGS THE GATES CAUGHT BEFORE THIS SHIPPED — all in my own tool

1. LEAVING OVERFLOW IN PLACE IS A CROSS-ORBIT TRIGGER. Once ^42 becomes d2's button,
   any surviving d1 reference means pressing d2's gate also fires d1's kick variant.
   Measured, not theorised: perfect.tidal ended up with ^42 driving BOTH d1 and d2.
   A new gate now asserts no button drives two orbits — 0 across the setlist.

2. THE CLASH GUARD HAD A HOLE. A clashed control also stays put, so it too becomes an
   obstacle — but I computed clashes in the same sweep that consumed them. bombe_dj:
   d2 was left on ^43 by a clash while d3 was still planned to move ONTO ^43. Fixed by
   iterating pass 2 to a fixed point, growing the sticky set until a sweep finds
   nothing new, with a non-convergence bail-out.

3. COMMENTING A LINE ORPHANS ITS CLOSING PARENS. A chain step spans several lines, so
   commenting only the line carrying the CC left the body live and the `)` dangling —
   `perfect: parse error on input ')'`, and desire went from building to not building.
   The unit is a `$`-chain SEGMENT; it is expanded to one, then paren-balance-checked,
   and the tool refuses rather than emit broken Haskell.

4. MULTI-BLOCK ORBITS WERE SILENTLY DROPPED. perfect.tidal declares `d1` SIX times as
   section alternates. Keying the plan by orbit number meant each later block
   overwrote the earlier one's plan, so the first block's ^42 -> ^41 vanished — the
   real reason gate 1 kept failing after I "fixed" it. An orbit's control set is the
   UNION across its blocks. Recovered 3 more moves (perfect 7->8, electric_hammer 3->5).

VALIDATION
  no button drives 2+ orbits    0 tracks (was 1)
  silent-eval --seeded          OK, every declared orbit emits events
  build failures                4, identical to before — no new breakage (#93)
  surface-columns               82/82 aligned, 0 renumbers
  pvlint                        0 errors, 9 pre-existing warnings (#70, dup d5)
  pytest                        471 passed
  check-boot / -blocks          OK

9 overflow controls are now commented with a FIXME naming the orbit, its slots, and
the restore path. That is the honest total: the surface is fully aligned, and every
control that could not fit says so in the file rather than in someone's memory.