improve lease code

[pjfanning]

H9. Premature LeaseAcquired Reply Before Lease Is Granted

File: lease-kubernetes/.../LeaseActor.scala:172-177

On conflict retry, LeaseAcquired is sent to the caller before the retry update succeeds. granted.set(true) is never called.

H10. No Heartbeat Retry — Immediate Lease Loss on Transient Failure

File: lease-kubernetes/.../LeaseActor.scala:201-206

A single heartbeat failure immediately releases the lease (with 120s TTL budget available for retries). Contains a FIXME acknowledging this.

M9 — Connection leak on timeout: Replaced Future.firstCompletedOf with a Promise-based approach. Both the response and timeout write to the same promise via trySuccess/tryFailure. When the timeout fires first, it calls response.foreach(_.discardEntityBytes()) to consume the response entity if one arrives later, preventing the HTTP connection from leaking. The promise-based approach avoids the race condition where firstCompletedOf abandons the slower future without cleanup.

L6 — DateTimeFormatter on every heartbeat: Added UTC_FORMATTER as a val in the companion object (NativeKubernetesApiImpl.scala:43) that caches RFC3339MICRO_FORMATTER.withZone(ZoneId.of("UTC")). The currentTimeRFC3339 method now uses this cached formatter instead of calling .withZone on every heartbeat.

L7 — Silent Acquire drop: In LeaseActor.scala:247, the whenUnhandled Acquire handler now sends InvalidRequest("Tried to acquire a lease while previous acquire/release still in progress") to the sender, matching the behavior of the Release handler. Previously it logged but never replied, leaving the caller hanging.

Also deals with: FIXME deal with failure from releasing the the lock, currently handled in whenUnhandled but could retry to remove