fix(core): ref-count store event listener to fix owner-first close leak

[dpol1]

Purpose of the PR

close #3033

storeEventListenStatus had the same owner-first-close bug #3017 fixed
for graph cache listeners: non-owner close() dropped the tracking entry
and skipped unlisten() as a no-op, leaving the owner's listener
registered but untracked — leak + no store-event cache invalidation.

Main Changes

• CachedGraphTransaction: add StoreListenerHolder inner class
  (listener, provider, refCount) + STORE_EVENT_LISTENERS static
  registry; acquire/release via ConcurrentMap.compute() with
  provider-identity guard for graph close/reopen. TODO block removed.
  Drop write-only storeEventListener instance field — listener
  ownership moved to StoreListenerHolder, field was assigned but
  never read.
• GraphTransaction: remove storeEventListenStatus declaration
  and orphaned ConcurrentHashMap import.
• CachedGraphTransactionTest: delete
  restoreStoreListenerStatusForKnownTeardownBug workaround; repoint
  reflection helper to STORE_EVENT_LISTENERS; add 5 regression tests:
  non-owner close keeps listener registered, last close removes listener,
  owner-first close with real STORE_CLEAR fires through provider and
  surviving tx clears cache, reopen re-registers a fresh holder. Provider
  is pooled across reopen — provider-replacement branch stays defensive,
  not exercised.

CachedSchemaTransaction* unaffected — per-instance, balanced 1:1.

Breaking change (internal API)

This removes the protected static final ConcurrentHashMap<String, Boolean> storeEventListenStatus field from GraphTransaction. The removal is
intentional and mirrors #3017, which already removed the sibling
graphCacheListenStatus (same protected static, same 1.7.0 vintage) with no
compatibility shim when graph-cache listening moved to the ref-counted holder.

• Scope: internal listen-tracking state consumed only by the in-tree
  CachedGraphTransaction; no external extension point uses it.
• No deprecated shim: a @Deprecated no-op field cannot preserve the old
  per-graph boolean semantics (now replaced by ref-counting), so it would be
  dead, misleading static state.
• Migration: none required unless external code subclasses
  GraphTransaction and reads this internal field directly. Store-event
  listening is now handled transparently by STORE_EVENT_LISTENERS.

Verifying these changes

• Need tests and can be verified as follows:
  • RED/GREEN TDD: new tests failed on master (NoSuchFieldException on
    STORE_EVENT_LISTENERS), green after fix.
  • 13/13 pass: CachedGraphTransactionTest
  • 36/36 pass: CachedGraphTransactionTest, CachedSchemaTransactionTest,
    CacheManagerTest, CacheTest

Does this PR potentially affect the following parts?

• Other affects (typed here): removes the protected static
storeEventListenStatus field from GraphTransaction — an intentional,
  internal-only breaking API change (see "Breaking change (internal API)"
  above), mirroring fix(server): align cache event actions in legacy EventHub path #3017. No known external use.

Documentation Status

• Doc - No Need

[imbajin]

Blocking: no. Summary: Please add one store-listener lifecycle test for the owner-first close path. Evidence: static review of CachedGraphTransactionTest; local Maven test was blocked by unresolved ${revision} dependency while current CI checks pass.

[imbajin]

Blocking: no. Summary: No obvious issues found in the current head. Evidence: CachedGraphTransactionTest passed locally and latest-head CI checks are green.

[dpol1]

@imbajin This PR covers the CGT fix. A follow-up improvement for CachedSchemaTransaction will be tracked separately. Not a bug or a hurry issue, but an improvement that will be calmly implemented.

[VGalaxies]

Review summary

• Blocking: no
• Summary: One low-severity compatibility risk remains from removing a protected field.
• Evidence:
  • git diff --unified=0 origin/master...HEAD -- hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/backend/tx/GraphTransaction.java

[dpol1]

Addressed - Also updated PR description

[imbajin]

Blocking: no. Summary: No obvious issues found in the current head. Evidence: git diff --check; CachedGraphTransactionTest passed locally (13 tests); latest-head checks pass.