feat(storage): persist estimated difficulty evidence

[hubab1]

Summary

• append additive SwiftData schema V4 with separate estimated-difficulty metric revisions and exact top-result evidence
• persist explicit estimated/unavailable state, heuristic and algorithm provenance, requested scope, actual ranking source, source fetch/computation times, explanatory notes, and structured redacted fallback classifications
• select immutable revisions deterministically so delayed contexts cannot regress visible results; compact visible losing revisions on later writes
• validate coherent score/confidence/evidence shapes, cap the logical current evidence set at ten, derive staleness from ranking fetch time, and keep missing distinct from unavailable

Safety boundary

• leaves released KeywordDailyMetric.difficultyScore and AppKeywordStats.difficultyScore untouched and performs no legacy backfill
• contains no ranking refresh, UI, MCP DTO, CSV, import, or provider wiring; those remain in 20c
• concurrent contexts may temporarily leave immutable losing revisions, but readers select one coherent winner and the next visible write compacts them
• 20c must guard delete-versus-in-flight persistence with query-existence, generation, and cancellation checks
• schema-changing draft: do not launch against the live user store before manual migration review

Validation

• 39/39 focused persistence, migration, and dependency tests
• 332/332 complete OpenASO tests
• released v0.3.2 fixture migrates through V4 with both legacy difficulty sentinels still 42, both new tables initially empty, and bundled artifact hashes unchanged
• direct V2-to-V4 and V3-to-V4 preservation tests
• reversed-save tests for existing rows and concurrent first inserts
• signed Debug build and codesign --verify --deep --strict
• three independent final reviews with no actionable P1/P2 findings after concurrency hardening
• app intentionally not launched because this branch changes the persistent schema

Dependencies and merge gate

This draft is stacked on #37. The eventual 20c integration also depends on #35 and #36. Keep this draft unmerged until the real-store migration/integrity review is complete.

Attribution

The broader estimated-difficulty idea was inspired by the upstream WIP commit akshaynexus@5de98ea. This persistence design and its tests are an independent rewrite; source link only, with no co-author trailer.