[P1] What #638's matching-side state count does to #636 — without funding #636 #664

Parent: #650. Read-only of PR #648 / #638. No new enumeration unless the control Catalan(w) is missing, and then only w≤6 on the Mac (NEED_HUAWEI for anything heavier).

#637 said the matching-side boundary state space is the cost-model risk for #636. #638 asked to enumerate it. PR #648 exists (Catalan for bonds, site frontier a strict subset, NNN noncrossing through w=8 — verify that sentence against the PR, do not take this parenthetical as the result).

Question

Write the cost-model implication in one notes file:

1. Quote the actual counts from PR probe638: boundary-state enumeration — Catalan holds for bonds, site frontier is a strict subset, NNN noncrossing through w=8 #648 (primal NN control vs matching NN+NNN, w=2..8).
2. Tripwire: NN control = Catalan(w), or the enumerator is unused.
3. One sentence: does this raise, leave, or lower [P1 reserve] Source-visible topology and all-width closure beyond the completed width-four laboratory #636's width ceiling relative to Catalan(w)?
4. Explicitly do not fund [P1 reserve] Source-visible topology and all-width closure beyond the completed width-four laboratory #636. If you think a TM is now cheaper or more expensive, that is a recommendation in the note, not a STATUS promotion.

Draft PR against main. No merge, do not close #636/#638.

Wait for ASSIGNED_MACHINE: local.

Related: #636 (not released), #637, #638, #648, #635.