Crash-safe frontier store (Badger/Bolt) & resumable jobs

## Summary
Provide a pluggable, persistent frontier and seen-set so long crawls can resume after crashes or restarts.

## Motivation
- Reliability for multi-day jobs
- Lower revisit rates after restart

## Scope
- Frontier interface remains the same
- Implement `badger` or `bolt` backend
- `--state-dir` config; `resume ` CLI

## Acceptance Criteria
- Kill the process mid-crawl; on resume, crawl continues with minimal revisits
- State file size and write rate documented

## Tasks
- [ ] Persistent frontier implementation + tests
- [ ] Job IDs and checkpoints
- [ ] CLI: resume command and flags