test: record doctor FTS feasibility

[ferueda]

Summary

• add an opt-in, provider-free measurement of document-ID-bounded FTS vocabulary reads used by the proposed doctor optimization
• record that production-shaped interval scans multiply the dominant work and reject that design
• keep production doctor behavior unchanged
• retire the rejected doctor-plan sections and add an evidence-gated single-pass feasibility plan
• preserve the independently executable orphan-maintenance work

Why

The proposed bounded-memory doctor design depended on document-filtered fts5vocab reads being cheap. PR #77 was stacked on this PR and merged into its branch, so this PR now contains both the original planning gate and the completed measurement.

The measurement found that 41 production-shaped intervals made total work 3.80× slower and grouped term-summary work 6.82× slower on the large generated cohort. The query-plan and scaling evidence is consistent with repeated whole-vocabulary traversal. That makes the proposed interval strategy unsuitable for production doctor.

Impact

No production src/ behavior, public CLI contract, schema, provider access, or retained-library state changes.

pnpm measure:doctor is an opt-in contributor command using generated data. It records exact equality, immutable-file, privacy, cleanup, timing, RSS, and query-plan evidence. The durable research note records the rejected approach so it is not repeated.

The successor plan evaluates a single-pass actual/expected comparison. It remains a feasibility gate and does not claim bounded total memory or authorize production changes.

Verification

• pnpm measure:doctor completed three generated runs with every hard correctness gate passing
• pnpm test test/doctor-measurement.test.ts
• pnpm check
• CI passed on macOS, Ubuntu, and Windows
• CodeQL passed
• independent code, test, measurement-method, and plan reviews passed