fix: confirmed slot gap during startup causes unrecoverable crash-loop

[gamandeepsingh]

fix: confirmed slot gap during startup causes unrecoverable crash-loop

On large databases the post-snapshot pipeline (dedup, index creation, cleanup) runs for several hours with finalization paused. During that window the gRPC buffer fills up, back-pressures the stream, and the upstream drops a slot — triggering a confirmed gap. check_slot_gap then paused the finalizer, which is the very worker that drives startup to completion, creating a deadlock. fill_gaps detected startup wasn't done and panic!'d, restarting the entire process and re-downloading the full snapshot. On a ~108 GB mainnet index this loop is unrecoverable.

Fix:

• SelfHealingState now holds a reference to snapshot_processing_state. In check_slot_gap, the finalizer is only paused if startup is already finished — gaps detected during startup are still queued for repair but the finalizer keeps running.
• In fill_gaps, the panic! is replaced with continue. The loop retries every 30s until startup finishes, then pauses the finalizer and repairs the queued gaps through the normal path.

Closes: #19