[Phase C1] Replace JSON store with SQLite-backed schema (extends #14)

[terisuke]

Tracked under ADR 0002 — Phase C. Detail: implementation plan §C1. Subsumes Issue #14.

Problem

The flat data/workflow_store.json snapshot has no versioning, no per-persona isolation, no Project/Article concept, and no way to fork sessions on edit. It is a process-global mutex-protected map that gets serialized as one blob (internal/infrastructure/repository/memory/workflow.go). Every Phase B and Phase A2/A3 feature needs durable, queryable history. Continuing on JSON forces a second migration later.

Scope

• New package internal/infrastructure/repository/sqlite using modernc.org/sqlite (pure Go, no CGO) — preferred — or mattn/go-sqlite3 if CGO is acceptable. Decision documented in PR.
• Schema migrations under internal/infrastructure/repository/sqlite/migrations/0001_*.sql, applied at boot via a tiny in-process migrator.
• Tables (minimum):
  • personas
  • author_sources
  • writing_style_guides (versioned per persona)
  • projects
  • articles
  • brief_sessions
  • brief_answers (with parent_answer_id for fork-on-edit)
  • drafts (versioned per article)
• persona_id is NOT NULL on every guide and answer to prevent cross-persona contamination.
• Default DB path: data/note_maker.db (gitignored).
• First-boot importer: if the JSON file exists, import it and rename to data/workflow_store.json.imported.{timestamp}.

Acceptance criteria

• All repository interfaces have SQLite implementations. Existing in-memory implementations remain available for tests.
• go test ./... passes against both the in-memory and SQLite implementations.
• Re-opening the app after a restart shows past projects, sessions, and drafts.
• Cross-persona guarantee: a unit test asserts that selecting persona A never returns guide rows from persona B.
• Migration test: a JSON store from before this change is imported cleanly.
• Fork-on-edit (Phase A1) is supported by brief_answers.parent_answer_id.

Out of scope

• Multi-user / multi-tenant (still single-user local).
• Cloud sync.
• A web admin UI for the DB (CLI-only inspection is enough).

Closes

• #14 (subsumed)

[terisuke]

Planning update after #22/#25/#56: keep #26 as the next product implementation cut, but include the new persisted surfaces in the SQLite schema/design. In addition to sessions/guides/briefs/drafts, persist source selector snapshots, source fetch metadata/content hashes, the composed question template version used by each session, draft evaluation/final-verification results, and section-regeneration versions. This is the persistence prerequisite before the full expensive Evo X2 media-matrix run becomes product history rather than loose files.

[terisuke]

Closed by PR #60 merged into develop. This adds the opt-in SQLite workflow store, migrations, restart tests, source snapshots, draft/verification persistence, section-regeneration versions, and web-app selection via WORKFLOW_STORE_DRIVER=sqlite. #14 remains open as the broader queryable-history/import-export umbrella.