CASSSIDECAR-454: Sidecar should wake up immediately on the instance r…

[mansikhara]

Summary

UpdateRestoreJobHandler now immediately notifies the restore system after writing a phase signal (STAGE_READY or IMPORT_READY) to the DB, eliminating the 5–10 minute idle wait for the next RestoreJobDiscoverer polling cycle.

• IMPORT_READY: calls RestoreJobManagerGroup.updateRestoreJob() to propagate the new status to in-memory trackers, allowing RestoreProcessor to unblock staged
  ranges on its next 1-second tick.
• STAGE_READY: calls RestoreJobDiscoverer.processJobNow() which discovers slices from the DB and submits them to RestoreProcessor immediately.

Design decisions

• Notification is fire-and-forget on a worker thread (executeBlocking) — does not block the event loop or delay the HTTP 200 response.
• Re-reads the full job from DB inside notifyPhaseSignalMaybe because RestoreJobDatabaseAccessor.update() returns a sparse object missing fields like keyspace and
  consistency level.
• Both calls are additive and safe to miss — the DB write remains the durable source of truth, and the discovery loop still picks up the signal on its next cycle if
  the immediate notification fails.

Test plan

• UpdateRestoreJobHandlerTest — new tests verify IMPORT_READY triggers updateRestoreJob and STAGE_READY triggers processJobNow
• RestoreJobDiscovererPhaseSignalIntTest — new integration tests verify ranges are created immediately on STAGE_READY without explicit discovery, and
  IMPORT_READY does not create duplicates
• Existing RestoreJobDiscovererNode*IntTest tests pass (regression)

[yifan-c]

CheckStyle is failing. There is a gradle task to run the check at local quickly.

./gradlew codeCheckTasks

[sarankk]

Thanks Mansi, looks good. Left some comments.