docs: mark P2/P3 done, and P4 with its measured per-repo state

[hunterdsp]

The phase checklist still showed P2–P4 unticked, so the plan understated progress while overstating what is left.

• P2 ✅ — build: adopt the cross-org Makefile standard (P2) doppler-dsp/doppler#558 plus four follow-ups. 50 hand-maintained targets → 0 shared + 25 local; help 30-of-50 (advertising two that did not work) → 67-of-67 generated. Entry records the two config-surface gaps doppler could not express (docs-check cannot accumulate; the build step has no flag surface, so an adopting C repo silently drops to one job) and the adoption-by-copy trap that left doppler unguarded for a day.
• P3 ✅ — skills://makefile-convention rewritten alongside #558, plus four sibling skills that named the renamed targets.
• P4 ⬜ — in progress, with measurements instead of a guess:

repo	run: steps	make	plumbing	unclassified
doppler	74	29	39	6
just-makeit	71	5	24	42

doppler's remaining 6 are 3 CI-native and 3 release-path (deliberately deferred — those steps only run during a release, so an error is invisible until it is expensive). just-makeit's 42 are mostly artifact.yml scaffold-smoke steps driving the jm CLI, which is the per-repo judgement P4 asks for rather than an omission.

Also recorded: two inline duplications P4 surfaced that a target-count audit misses — docs.yml reimplementing make gen-c-api and already diverging from it (uv run mkdocs vs uv run --group docs mkdocs; a cp that merged where the target replaces, so a deleted C symbol's page survived forever in the deployed site), and make test-python selecting a different test set from CI's pytest — the RFC's own complaint surviving the port meant to end it.