Preserve observation history across repeated fetch runs

[CrisSTEM]

Summary

• preserve append-only observation history across repeated fetch runs
• add a non-destructive storage.SyncPack path for fetch-time catalog/config synchronization
• keep seed-demo as the explicit destructive reseed path
• add CLI and storage tests covering repeated same-DB runs and multi-source accumulation
• update runtime/docs language to reflect the new history-preserving fetch behavior

Testing

• go test ./...
• make ci

Notes

• fetch now bootstraps the database and synchronizes the selected config pack before execution without clearing prior observation history
• repeated same-DB runs preserve prior fetch_runs, append-only metric_snapshots, and deduplicated content_items
• seed-demo remains the explicit catalog/bootstrap reseed path

Deferred

• destructive catalog cleanup/garbage collection is still not part of the fetch path
• scheduler execution, alert materialization, candidate discovery, API, and frontend work remain deferred

Fixes #45