[Fix][operator] Fix volcano podgroup stuck in inqueue state after rayjob completes

[fangyinc]

Why are these changes needed?

When a RayJob completes (SUCCEEDED/FAILED), the associated RayCluster is deleted (when shutdownAfterJobFinishes: true), but the Volcano PodGroup remains stuck in Inqueue state. This causes queue resources to remain occupied even though the job has finished, preventing new jobs from being scheduled.

Root Cause

KubeRay never implemented cleanup logic for Volcano PodGroups. When RayJob completes:

1. RayCluster is deleted
2. PodGroup persists (OwnerReference points to RayJob, not RayCluster)
3. Volcano scheduler continuously recalculates PodGroup status based on pod counts
4. With 0 running pods, the status gets reset to Pending → Inqueue by Volcano's control loop

Solution

Added a CleanupOnCompletion() method to the BatchScheduler interface that deletes the PodGroup when RayJob reaches terminal state. Deleting is necessary because marking as Completed doesn't work - Volcano's scheduler overrides the status in its next cycle.

Related issue number

Closes #4473

Checks

• I've made sure the tests are passing.
• Testing Strategy
  • Unit tests
  • Manual tests
  • This PR is not tested :(

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

[fangyinc]

@Future-Outlier @andrewsykim PTAL. Thanks you~

The failed CI does not seem to be caused by this PR.