Core: Make InMemoryLockManager heartbeat map thread-safe

[LuciferYang]

InMemoryLockManager.HEARTBEATS is a static map shared by every lock manager in the JVM and is written whenever a lock is acquired or released. The sibling LOCKS map is already a ConcurrentMap, but HEARTBEATS was left as a plain HashMap, so threads acquiring or releasing locks on different entities mutate it without synchronization. That can corrupt the map and leak the scheduled heartbeat futures it tracks.

This makes HEARTBEATS a ConcurrentMap and replaces the non-atomic containsKey/remove cleanup in acquireOnce with a single atomic remove.

Tests:

• distinct entities: many rounds of concurrent acquire/release across separate keys, asserting no heartbeat is left behind;
• shared entity: 16 threads contend on one key, asserting at most one heartbeat while it is held and none after release;
• takeover: a lock whose holder never releases is re-acquired by a new owner once it expires, asserting the stale heartbeat is cancelled and replaced. This is the only path that reaches the non-null cleanup branch the change touches.

[wombatu-kun]

This atomic remove cancels a prior heartbeat only when one is still tracked for the entity - the orphaned/expired-takeover case where the holder never released. Both new tests acquire-then-release each iteration, so release clears the heartbeat before the next acquire and this remove returns null every time; no existing test hits the non-null path either. The branch this PR actually changes is therefore unexercised. Consider a focused test that leaves a heartbeat orphaned (lock expires or its heartbeat task dies without a release), then re-acquires and asserts the stale future is cancelled and replaced.

[LuciferYang]

Good catch, you're right. release() always removes the heartbeat first, so both new tests only ever hit the null path and the branch this PR changes stayed uncovered.

I added testAcquireCancelsOrphanedHeartbeatOnTakeover: it acquires without releasing, lets the lock expire (long heartbeat interval so it isn't renewed), then re-acquires with a new owner. That reaches the non-null branch, and the test asserts the stale future is cancelled and replaced by a fresh one. Needed a small @VisibleForTesting heartbeat(entityId) accessor to check the future identity. Thanks for the review!