Re-derive v5 coverage plan against current main (supersedes #153)

[millsmillsymills]

#153 (the >=90% coverage plan) was closed unmerged from the review/merge cycle. This issue carries forward its still-useful research and the work needed to make it mergeable.

Why #153 was closed

• Its Phase 1 (read-backfill) already shipped via feat(domain): read-backfill GET tools #173 / feat(livedns): read-backfill GET tools #174 / feat(cert,email): read-backfill GET tools #175 — main is now 91 tools (54 read / 29 write / 8 purchase), so the plan's "71 tools" baseline is stale.
• The plan's headline math was internally inconsistent (per-area covered counts, "all 81" vs "35 of 55" vs the enumerated Phase 4/5 checklists, 71-vs-72 baseline).
• The 208-endpoint denominator was hand-counted from live RAML and is not reproducible from the repo — there is no vendored spec to diff against, and §5 itself notes it will drift. Re-baselining the percentages would propagate unverifiable numbers.

What to do for a mergeable v2

1. Build the programmatic denominator first (the §5 follow-up): add a scripts/ RAML-snapshot fetcher that captures the live /v5 endpoint set, vendored into the repo so tests/unit/test_tool_schema_matrix.py can guard real coverage, not just self-consistency.
2. Re-derive per-area covered counts from the generated matrix (authoritative), not by hand. Current registered surface: billing 3, certificate 8, domain 38, email 16, livedns 21, organization 5.
3. Mark Phase 1 done; re-scope Phases 2–5 against the new baseline; reconcile the implement/defer headline against the enumerated per-area tables so every total/percentage ties out.

Worth preserving from #153 (verbatim research, re-confirm against fresh RAML)

• Appendix A endpoint inventories for the three newly-discovered zero-coverage areas: mailbox (18), linkedzone (12), comment (3).
• The two CLAUDE.md "v5 gap" reconciliations: PATCH /v5/domain/domains/{domain}/status (transfer-lock) and POST …/transferout appear in live RAML but are documented as gaps — verify against a real account before building.
• Phase structure (read → free-write → gated-purchase, per area) and the purchase-tool default-off gating rules.

The #153 branch (docs/tool-coverage-90pct-plan) holds the original draft if needed.