fix(surface): the helper-CC scan was counting CCs that appear only in COMMENTS

surface-columns reported 11 CCs as "held by BootTidal" — including 18, 34 and 77.
All three are false. Every one appears only inside a `--` comment:

    BootTidal.hs:299   --   midiOn ("^34" - "^18")   (perfect.tidal:63)
    BootTidal.hs:322   -- midiGGlobal used to read the LIVE fader: orDef 0.769 "^77" * 1.3

one documenting a track's idiom, one a note about a retired helper. BootTidal
claims none of them.

This is the dangerous direction of wrong. An INFLATED helper set says "these
controls are unavailable", so it silently shrinks the design space for the #92
column migration — B-row column 6 (^34) and A-row column 6 (^18) would have
looked spoken-for when they are free, and ^77 would have looked like a Tidal-side
conflict with Ardour's newly-learned fader 1 when there is none. A wrong number
that closes doors is worse than one that opens them, because nobody goes looking.

Caught it by grepping BootTidal for the three CCs while planning #92 and finding
every hit was a comment line. Fix is one line: run strip_comment over the helper
block before matching, the same way pvlint and silent-eval already do.

Corrected numbers: 8 helper CCs (41, 49, 50, 51, 73, 74, 75, 93) — gMask, the
three DJ filters, the three mutes, and panic. Exactly the set you would predict
from reading the helpers, which is the tell that it is right this time.

Knock-on: the renumber count rises 174 -> 192, because ^34 and ^18 are no longer
excused as immovable helpers and now correctly count as track controls that would
have to move. #92 updated.

tools/surface-columns.py

@@ -91,6 +91,14 @@ def helper_ccs(boot: Path = BOOT) -> set[int]:
     if i >= 0:
         j = text.find("\n:}", i)
         text = text[i:j if j > 0 else len(text)]
+    # STRIP COMMENTS. Without this the scan reported 11 helper CCs including 18,
+    # 34 and 77 -- all three of which appear ONLY inside `--` comments
+    # (`midiOn ("^34" - "^18")` documenting perfect.tidal, and a note about a
+    # retired midiGGlobal that used to read "^77"). BootTidal claims none of
+    # them. An inflated helper set is the dangerous direction of wrong: it would
+    # have told us three usable controls were unavailable, and quietly shrunk
+    # the design space for #92. A parser miss must never look like a data fact.
+    text = "\n".join(strip_comment(l) for l in text.splitlines())
     return {int(m) for m in CC_REF.findall(text)}