docs: correct six stale comments found by a fact-check pass

[mcdope]

Six comments that describe the code as it was rather than as it is. No behaviour changes — comments, one doc paragraph, and one config comment.

They surfaced because three independent agents fact-checked a draft of some launch copy against the repo. Nobody goes looking for stale comments; they turn up when something forces a close read.

What was wrong

Where	Claimed	Actually
lore.ts:21-23	HP_PER_COMPLEXITY is 25	35
enemies.ts:412-415	room total is 25 * complexity, Edge Case roll is 10-15	35, and 25-35
mapGenerator.ts:263	Edge Cases are never in a normal room	set in three places since 2026-08-20
main.ts:327	third tab is "GitHub"	it is "Repo", and takes GitLab/Codeberg too
decisions.md	assertAllRoomsReachable is a "dev-time assertion"	runs in production, and logs rather than throws
vitest.config.ts:100	coverage is 99.94/99.79/99.62/99.97%	99.9083/99.5225/99.84/99.99%

Notes on the fixes

The two HP comments now name the constants instead of restating their values, so this particular drift cannot recur. The two deliberate historical references to 25 — in weapons.ts and combatConstants.ts, both describing the 25 → 35 change itself — are left alone.

mapGenerator.ts kept the half that is still true. spawnEdgeCaseEnemies does populate the corridor features exclusively and is the only thing that populates them; it is the converse ("and normal enemies never spawn in one") that stopped holding. types.ts already carried the corrected wording and is now cross-referenced.

The coverage figures are restated in units rather than percentages, because that is the part worth knowing: statements are at 13,073/13,085 and branches at 5,627/5,654, so each can afford exactly one more uncovered unit before the gate fails. The thresholds no longer "sit just below" what is measured — statements is essentially on the line, and that is worth a reader knowing before they wonder why an unrelated PR went red.

That last number is itself a correction: the first draft said "a single new uncovered statement fails the gate", from a measurement taken on a branch that carried extra code. The arithmetic says two.

Also in here

One unrelated commit, acd3acf, adds pr/ to .gitignore — a local scratch directory for launch copy and screenshots. It has to be ignored rather than merely untracked, because run-balancing-capture.mjs refuses to start on any non-empty git status --porcelain.

Verification

• npm run build — clean, bundle hygiene clean
• npx vitest run --dir src — 3,682 tests, 135 files, all passing
• npx vitest run scripts/lib/docPins.test.mjs — 70 passing
• Coverage re-measured on this branch: 99.9/99.52/99.84/99.99, all four gates clear

🤖 Generated with Claude Code

https://claude.ai/code/session_017ncJfux8GDacSTeLDcqrhr