docs: stale inventory counts and snippet-length claim outside README escape validate-counts

[TMHSDigital]

Why it matters

validate-counts only checks substrings in README.md, so the same inventory numbers repeated in other docs drift without anyone noticing. Agents read AGENTS.md first, and it currently gives them the wrong showcase count.

Evidence

Location	Says	Actual
AGENTS.md:23-24	"59 examples, and 26 showcase pieces (counts are CI-enforced against README.md)"	36 showcase pieces. The line also has a stray ).
CONTRIBUTING.md:310	"26 showcase pieces" (as the model example)	36
ROADMAP.md:259	"26 of 42 examples pass all floors" (measured 2026-07-24)	59 examples now. Re-measure, or date-label it as historical.
README.md:1264, CLAUDE.md:99, CONTRIBUTING.md:51,100	Snippets are "5 to 50 lines each"	snippets/export_preset_unreal.py is 67 lines, snippets/lod_chain.py 55, snippets/action-ensure-channelbag-for-slot.py 52

Suggested approach

• Fix the numbers above.
• Extend validate-counts to scan AGENTS.md and CONTRIBUTING.md for the same N showcase / N examples patterns. Better still, drop hard counts from the secondary docs and point them to the README.
• Add a snippet line-count check to Validate (5 ≤ lines ≤ 50). Then either trim the three offenders (e.g. split export_preset_unreal.py into glTF and FBX halves) or raise the documented cap.

Done when

• Every inventory number in AGENTS.md, CONTRIBUTING.md, README.md and CLAUDE.md matches disk and is CI-checked.
• The snippet-length claim is true and enforced.