Add DB-backed cleanup runs and cleanup items

[chubes4]

Goal

Replace filesystem cleanup plan workflows with DB-backed cleanup runs/items.

Scope

Add tables/repositories for:

• datamachine_code_cleanup_runs
• datamachine_code_cleanup_items

Required workflow

• workspace cleanup plan --mode=<mode> creates a cleanup run and item rows.
• workspace cleanup apply <run-id> schedules or applies rows by querying item state.
• workspace cleanup status <run-id> aggregates item/job state.
• workspace cleanup evidence <run-id> reads run/item/job evidence.
• workspace cleanup resume <run-id> resumes pending/failed chunks.
• workspace cleanup cancel <run-id> cancels pending items/jobs.

Data model expectations

Each cleanup item stores:

• handle/worktree reference
• planned action
• current status
• reason_code/reason
• chunk/job ids
• bytes reclaimed
• bounded evidence JSON

Migration from current behavior

• File-based --apply-plan=/path.json becomes an escape hatch only.
• CLI help should stop documenting redirect-to-file as the normal path.
• Data Machine job engine_data can store chunk-local payloads, but run/item state belongs in DMC tables.

Tests

• Plan creates run/items.
• Apply claims pending items and updates status.
• Status/evidence aggregate item and child job state.
• Retrying a run is idempotent.
• No filesystem plan files are created.