perf(indexer): cap aggregate score update cadence

[raymondjacobson]

Summary

• add a 10 minute pause between full aggregate score update passes
• keep shutdown responsive to context cancellation

Evidence

• production read-replica pg_stat_statements showed the aggregate score batch query as the top total-time read-path statement: ~39.5M calls, ~13.4M total seconds, ~340ms mean
• live pg_stat_activity showed core-indexer actively running the same user-score batch query after the read replica drain
• the old loop immediately started another full pass after each full scan, so this bounds repeated read load without changing scoring semantics

Test

• go test ./indexer ./jobs -count=1