docs(roadmap): retire three cliff TODOs that measurement has falsified

[wormeyman]

Bookkeeping half of the issue #16-#24 audit. No code changes.

docs/noise/client-preview-ROADMAP.md is what someone reads before picking up cliff work, and it still listed three things as open that are not - one of them a to-do for a rule the game does not have.

roadmap line	was	now
"Ore excluded from cliffs"	open, low priority	withdrawn - no such rule exists
"Still deferred: fixImpossibleCells"	deferred	ported (PR #32, dea73ac), changes zero predictions
"Still deferred: wouldCollide"	deferred	moot - no Nauvis cliff touches water

Each was verified against the repo before editing, not taken from a summary: cliffPlacement.ts:105 and test/cliffFixImpossibleCells.spec.ts exist; dea73ac merged 2026-07-28T17:36Z.

The Nauvis residual itself stays open. PR #57 reclassified it from a field error to a rule error, and the roadmap now says that rather than pointing at three retired causes.

Done alongside this, on GitHub

• Vulcanus cliffs sit on ore at the chance rate (16.7%); the game separates them to 0.2% #24 closed as not planned, per its own last comment. generateCliffs() (0x1016229b4) touches no resource data, so cause 1 had nothing to implement; cause 2 re-points at Vulcanus cliffs: 12.5% of matched cells get the wrong cliff_orientation (residual after the placement-count fixes) #18. Its sharpest artifact - 47 cliff cells our rule places inside entirely-calcite footprints where the game places 0 - was carried over to Vulcanus cliffs: 12.5% of matched cells get the wrong cliff_orientation (residual after the placement-count fixes) #18 rather than lost.
• Vulcanus cliffs: 12.5% of matched cells get the wrong cliff_orientation (residual after the placement-count fixes) #18 retitled. "have no entity-level validation" has been false since PR Validate Vulcanus cliffs against real entities: 57-69% recall, 1.5x over-placement #20 landed that fixture; the title now names the actual defect. A comment records the current figures and the next experiment.
• Register: known approximations and unpinned semantics in the placement/cliff port (audited 2026-07-30: 2 of 6 already done) #22 audited. Items 2 and 6 were already done - item 6 landed 37 minutes after the issue was filed - and the body had never been edited. Both are struck through with evidence, and item 1 is updated: the rock huge/big split has since generalised to Nauvis (region 0: 42 huge / 149 big = 22% huge where argmax predicts 0%), making it a shared root cause with Cross-overlay occupancy: earlier autoplacers block later ones (enemy bases are 47% over in region 0) #16.

This is the failure mode documented-rationales-expire describes: a register whose entries are wrong is worse than no register.