Improve indexer resilience for Redis outages, branch renames, and deleted repos

[aanogueira]

The indexer would hang indefinitely when Redis became unreachable, silently mark repos as empty when their default branch was renamed, and waste retries on repos deleted from the code host.

Redis outage recovery:

• Worker exits after ~5 min of consecutive Redis failures so K8s restarts
• CleanupOldJobs now removes ghost entries from running/pending status indexes, type indexes, and processing set
• DequeueForShard cleans up ghost job index entries on pop

Branch rename detection:

• processIndexJob auto-detects the actual default branch when configured branch doesn't exist (prefers main > master > first available)
• Updates the database so future jobs use the correct branch

Deleted repo detection:

• Add PermanentError type to skip retries on unrecoverable failures
• Detect 404/not-found git errors during clone/fetch
• Mark deleted repos as excluded immediately instead of retrying 3 times
• New repo_not_found category in failure metrics

Other:

• Set fsGroupChangePolicy: OnRootMismatch to speed up pod restarts
• golines formatting fixes