Add failing tests proving premature LeaseAcquired bug (#3402)

[Aaronontheweb]

Summary

Adds deterministic reproduction tests for the premature LeaseAcquired split-brain bug in both Azure and Kubernetes LeaseActor implementations.

These tests intentionally FAIL to prove the bug exists. The fix will come in a subsequent commit.

The Bug

When a CAS conflict occurs during the Granting state and the blob/configmap has no owner (version moved on), LeaseActor sends LeaseAcquired to the caller before the retry write completes (K8s line 365, Azure line 362). If the retry then fails (another node takes the lease), the caller already believes it holds the lease — but it doesn't. localGranted is never set to true, heartbeat never starts.

Test Results

Test	K8s	Azure	Expected
ShouldNotSendPrematureLeaseAcquiredWhenConflictRetryIsStolen	FAIL	FAIL	Will pass after fix
ShouldGrantLeaseOnlyAfterConflictRetrySucceeds	PASS	PASS	Happy-path verification

Failure Output

Expected a message of type LeaseTaken, but received LeaseAcquired

This directly proves the premature LeaseAcquired is sent before the retry write, confirming the split-brain scenario described in #3402.

All existing tests pass

• K8s: 47 passed, 1 skipped, 0 failed
• Azure: 45 passed, 1 skipped, 0 failed

Fixes #3402

Closes #3403

[Aaronontheweb]

Integrated Gregorius' original fix commits into this branch and extended them to Kubernetes for parity.\n\nIncluded with attribution via cherry-pick:\n- 611112c (Fix premature LeaseAcquired in LeaseActor Granting state)\n- a2ed0e2 (Harden test)\n\nAdditional commit in this branch:\n- Kubernetes equivalent fix for premature LeaseAcquired in the null-owner conflict retry path.\n\nThis keeps the repro-first record intact while preserving prior work and author credit.