test: measure doctor FTS feasibility

[ferueda]

Summary

• add an opt-in, provider-free doctor document-interval feasibility measure with strict worker isolation, immutable file checks, exact semantic equality, and failure cleanup
• record that production-shaped document-ID interval scans are 3.80× slower on the large generated cohort and accept rejection of that production design
• retire the rejected doctor plan sections and add an evidence-gated single-pass CPU feasibility plan without claiming bounded total memory

Why

The proposed bounded-memory design depends on document-filtered fts5vocab reads. Measurement showed those predicates retain the same virtual-table scan shape and multiply the dominant work by interval count. This PR keeps production doctor unchanged while preserving reproducible evidence and a safer next gate.

Impact

No production src/ behavior, CLI contract, provider access, or retained-library state changes. pnpm measure:doctor is an opt-in contributor command supported on macOS/Linux because its file-permission evidence requires POSIX modes.

This is stacked on #76. Merge #76 first, then retarget this PR to main if GitHub does not do so automatically.

Verification

• pnpm measure:doctor (three complete generated runs; exact hard gates passed)
• pnpm test test/doctor-measurement.test.ts
• pnpm check — 125 test files, 1,302 tests, build, dist smoke, and package smoke
• independent code, test, and measurement-method reviews passed