fix(imagecollector): prevent OOM cascade on controller restart

[syedazeez337]

The eraser-controller-manager creates multiple ImageJobs when it restarts, causing an OOM cascade where each ImageJob spawns pods on every node, overwhelming the API server and leading to further OOM events.

This fix adds synchronization to the 'first-reconcile' mechanism:

• Use mutex to prevent concurrent first-reconcile executions
• Track completion state to avoid redundant job creation
• Properly detect existing running jobs before cleanup
• Re-list jobs after cleanup for accurate state
• Only create new job if no running jobs exist

Testing confirms the fix:

• 10 concurrent controller restarts now create only 1 ImageJob (vs 10 before)
• No redundant pod creation
• No API server pressure
• No OOM cascade

Resolves: #1169