fix: uncordon nodes left cordoned after lock loss

[xavierleune]

Summary

• Fix nodes remaining cordoned indefinitely when kured loses the lock during reboot (e.g., lock TTL expiry, lock taken by another node)
• Detect orphaned cordoned nodes using --pre-reboot-node-labels or --annotate-nodes markers

Problem

When a node reboots and kured restarts, if lock.Holding() returns false (lock expired or released), the uncordon logic is skipped entirely. The node remains cordoned with pre-reboot labels but kured doesn't recognize it as "its own" node to uncordon.
This is particularly problematic with MicroOS or similar auto-updating OS where a second reboot may be required immediately after the first.

Fixes #63

Solution

After checking lock.Holding(), if we don't hold the lock, check if the node:

• Has the KuredRebootInProgressAnnotation (if --annotate-nodes is enabled), OR
• Has the --pre-reboot-node-labels labels

If either condition is true AND the node is unschedulable, uncordon it.