fix(tools-health): audit-cdecl is a HEALTH check, not a regression suite — sample by default

Drew, correctly: "this is a tools health test, not a full regression test."
audit-cdecl re-parsed every declaration in all 4,168 TUs and handed each to real
gcc — ~787s of pure-Python collection before the first cc1 call. It made
`make tools-health` unrunnable: >15 min, killed twice, never completed once.

`--limit` already existed and its own help calls it "a fast smoke run"; nothing
was using it. Sampled by default (CDECL_AUDIT_TUS ?= 60); the exhaustive form
stays as `audit-cdecl-full` for when cdecl.py itself changes.

  audit-cdecl : >9 min -> 61s (4,777 declarations adjudicated, 0 rejected)
  tools-health: never completed -> 333s, rc=0, all green

Known limit, recorded not hidden: --limit takes the FIRST N TUs, not a random
sample, so the smoke run always exercises the same files. Randomising the sample
(or rotating by seed) is the follow-up.

Makefile

@@ -184,7 +184,16 @@ audit-corpus:
 # so the candidate set is every depth-0 statement, and there is no hand-maintained candidate regex
 # to rot. The real cross-gcc then adjudicates BOTH the parse and the residue (R34): it compiles each
 # declaration beside this parser's reconstruction of it, and a statement gcc also rejects is not C.
+# SAMPLED BY DEFAULT (P31 S70). This is a HEALTH check, not a regression suite. The full pass
+# re-parses every declaration in all 4,168 TUs and hands each to real gcc: ~787s of pure-Python
+# collection before the first cc1 call, and it made `make tools-health` unrunnable (>15 min, killed
+# twice). The exhaustive form still exists as `audit-cdecl-full` — run it when cdecl.py itself
+# changes, not on every health check.
+CDECL_AUDIT_TUS ?= 60
 audit-cdecl:
+	$(VENV_PY) tools/cdecl.py --audit --gcc --limit $(CDECL_AUDIT_TUS)
+
+audit-cdecl-full:
 	$(VENV_PY) tools/cdecl.py --audit --gcc
 
 # Binary-citizenship gate (Phase-28 T7, R36 via R32). Asserts every onboarded binary