[state-prune snapshot] Resume/restart support (MVP + optional worklist persistence)

[jolestar]

Parent: #3858

Goal

Make long-running snapshot builds resilient to restarts.

MVP

• Restarting should safely continue without corrupting output.
• Preferably leverage output DB as visited set to skip already-written nodes.

Optional

• Persist worklist/counters periodically to snapshot_progress.json and resume from it.

Acceptance

• Kill mid-run and restart; completes and produces the same final snapshot as a clean run.
• Clear operator logs indicating resume/skip behavior.