Summary

Thank you for maintaining Crawlith — I ran into a small snapshot/export edge case while using Crawlith for SEO internal-link graphing and opened #103 with the reproduction details.

This PR scopes page loading for non-single snapshots to pages whose last_seen_snapshot_id matches the selected snapshot. That keeps graph exports aligned with the current crawl's normalization policy instead of including pages that were only seen in older snapshots.

Fixes #103.

Problem

If a site is first crawled with query strings preserved and later crawled with --no-query, the latest export can still include older query-URL nodes such as:

/book-a-call?intent=header-cta
/free-audit?intent=home-hero

The crawler is normalizing new discoveries correctly, but loadGraphFromSnapshot() ultimately relies on page repository snapshot queries that include pages first seen in older snapshots. That makes --no-query appear ineffective in exported graph nodes.

Changes

• Scope getPagesBySnapshot() to p.last_seen_snapshot_id = ? for non-single snapshots.
• Scope getPagesIteratorBySnapshot() the same way for graph loading/export.
• Scope getPagesIdentityBySnapshot() to the current snapshot so edge materialization uses the selected snapshot's page set.
• Preserve existing single snapshot behavior, which is still metrics-scoped.
• Add a DB-layer regression test covering stale query URL exclusion on a later snapshot.
• Add a changeset for @crawlith/core.

Validation

Ran locally:

pnpm run lint
pnpm test
pnpm build

Also verified manually against a real crawl workflow:

1. Created a dirty snapshot with query URLs preserved.
2. Reran the same site with --no-query without cleaning the DB.
3. Exported the latest graph.
4. Confirmed queryUrlNodes: 0 in the exported graph.json.

Thanks again — happy to adjust the query semantics if you'd prefer a different snapshot-scoping approach.