Add failing tests proving heartbeat surrender bug (#3404)

[Aaronontheweb]

Summary

• Adds deterministic reproduction tests for Bug Transient heartbeat failure immediately surrenders lease, causing unnecessary ~96s shard outage #3404 in both Azure and Kubernetes LeaseActor specs.
• The tests prove a transient heartbeat Status.Failure currently causes immediate lease surrender instead of retrying within the TTL safety window.
• These tests intentionally fail to establish proof before any fix is applied.

What these tests assert

• After successful acquisition and heartbeat start, a single transient heartbeat failure should not:
  • set Granted to false,
  • invoke leaseLostCallback, or
  • stop heartbeat retry attempts.
• Instead, actor should remain granted and schedule another heartbeat write attempt.

Current behavior (bug)

Both implementations immediately surrender lease on Status.Failure while in Granted:

• Azure: LeaseActor.cs in Granted handler (Status.Failure branch)
• Kubernetes: LeaseActor.cs in Granted handler (Status.Failure branch)

Observed failure in both test suites:

• Expected retry update message within heartbeat interval window
• Actual: no retry; actor transitions to idle and marks lease lost

Why this PR

Previous fix attempts went straight to implementation. This PR is repro-only to build a deterministic, reviewable proof that the bug is real in both providers before proposing fixes.

Fixes #3404

Closes #3405

[Aaronontheweb]

Integrated Gregorius' original Azure heartbeat-retry fix into this branch and extended it to Kubernetes for parity.\n\nIncluded with attribution via cherry-pick:\n- 0cf2b4e (Retry heartbeat on transient failure instead of immediately surrendering lease)\n\nAdditional commit in this branch:\n- Kubernetes heartbeat retry implementation + matching test updates.\n\nThis preserves the repro-first evidence trail while carrying forward prior work with proper credit.