Task: SQLite storage layer

Description

Create the SQLite storage layer for the quality monitoring system using bun:sqlite. This includes database initialization, schema creation, migrations, CRUD operations for responses and regressions, query helpers for aggregations, and automatic retention cleanup.

Acceptance Criteria

• packages/opencode/src/quality/store.ts created with full SQLite wrapper
• Database initialized at configurable path (default .opencode/quality.db)
• responses table with columns: id, session_id, model_id, provider_id, timestamp, prompt_hash, response_hash, composite_score, correctness_score, feedback_score, analysis_score, diagnostic_notes (JSON), is_collaborative, tokens_input, tokens_output, cost
• regressions table with columns: id, response_id, detected_at, score_delta, severity, component, description
• Indexes on: responses(session_id), responses(model_id), responses(timestamp), regressions(severity)
• CRUD operations: insertResponse, updateResponseScore, insertRegression, getResponsesBySession, getResponsesByModel, getRecentResponses(limit)
• Aggregation queries: getModelAverages, getRollingAverage(windowSize), getScoreTrend(days)
• Retention cleanup: deleteOlderThan(days) runs on DB init
• Schema version tracking for future migrations
• Graceful error handling — DB failures logged, never crash main workflow

Technical Details

• Use bun:sqlite (built-in, zero dependencies)
• Use Database class from bun:sqlite with WAL mode for performance
• Store DB path resolved relative to project directory via Instance.directory
• Follow existing Instance.state() pattern for per-project DB lifecycle
• Create packages/opencode/src/quality/schema.ts with Zod schemas for type safety on insert/query results

Dependencies

• None — this is a foundation task

Effort Estimate

• Size: M
• Hours: 6-8
• Parallel: true