feat(reconcile): surface indeterminate rows for review

[kevinatlee]

Summary

Surface every catalogue-reconciliation row kept because provider or profile evidence is incomplete. The response now separates informational indeterminateRows from removal candidates, and the modal renders each kept title with its provider and evidence reason. Closes #2892.

Safety invariant

Incomplete provider/profile evidence never becomes deletion evidence. Indeterminate rows remain absent from candidates, have no selection control, do not affect the removal count, and are never sent by the UI to Apply. Apply still recomputes candidates and intersects the requested IDs with the current candidate set, so submitting an indeterminate ID directly does not delete it.

Behavior before / after

Before, the API and modal exposed only summary.indeterminate, leaving users unable to audit which local rows were protected.

After, indeterminateRows reports the local book ID, title, metadata provider, and one of five display-only reasons:

• language_unknown
• language_evidence_lookup_failed
• edition_evidence_unavailable
• partial_catalogue
• unmatched_cross_provider

The existing deletion-candidate reasons and candidate algorithm remain separate and unchanged in substance.

Jenny Han reproduction

For the English-only Hardcover case described in #2892, the modal now makes the protected rows auditable from reconciliation evidence rather than asking users to infer them from titles. The implementation does not assume that the six named titles are the six protected rows; it reports whichever local rows actually correlate to incomplete language/provider evidence at preview time.

Checklist

• Tests added or updated
• Doc-update gate cleared (README.md, API docs, user guide, and changelog fragment)
• Wiki page updated for the user-facing behavior

Tests / verification

• Focused reconciliation Go tests
• go vet ./...
• golangci-lint run ./... (v2.11.4; 0 issues)
• cd web && npm run lint (0 errors; 7 pre-existing warnings)
• cd web && npm run typecheck
• cd web && npm run build
• cd web && npm test (105 files, 1,175 tests)
• git diff --check
• go test ./cmd/... ./internal/... — attempted on Windows; unrelated existing platform assumptions fail in telemetry/ABS/path, POSIX permission, symlink, and hardlink tests. The telemetry failure passed when rerun in isolation, and all focused reconciliation tests pass.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!