[toprameen]

Replace spec_of’s boolean arrays with declaration arity and the template prefix count. Checked templates have a contiguous leading prefix; datatypes and term heads keep prefix zero.

Local checks: repository gate and whitespace check pass. Exact BendTT text and out-of-scope results match on 331 checked books, including a focused fixture, with four repeated-call comparisons and 15 direct boundary checks. All 14 local BendTT verdict comparisons match the baseline; 13 CLI verdicts pass and five parser diagnostic pins match.

[Lorenzobattistela]

Thanks, this is a clean follow-up to #1203. Since #1203, spec_of was always a leading prefix of tld.x trues, so storing the count instead of the array is exact. It relies on two things the parser guarantees: a ~ only comes on leading clauses or binders, and a def that fills a law has tele.length >= law.x, so x <= n always holds.

I checked it merged onto main 53fc961, using main's own gate files:

• repo 50/50, ping 47/47, test 1577/1577
• perf 123/124, the same as main (a first run showed 120/124 from node contention; the rerun matches)
• gates/safe.ts tests: the table is byte-identical to main's (1018 agree, 0 timeouts; the one false reject, agree_32, is already on main)
• -o: every tests/*/*.bend emitted on both trees; all 1577 exit codes match and all 1090 .bendtt files are byte-identical

So the change has no effect on output. Two small nits, neither blocking:

1. sp is a boolean[] on main and a count after this PR (in args and in group_emit's rest). Code that still does sp[j] on a number gets undefined with no type error, and every parameter silently counts as unspecialized. A name like nx would make
2. The comment on the deleted spec_of was the only place safe.ts said which parameters get specialized. The new code also depends on those parameters one-line comment at root_cols or args(e.g. "a def's first x parameters, its ~ ones, are specialized") would keep that rule written down.