Extract concurrency utilities, add multithreaded tests, fix two races

[predatorray]

Pull the hand-rolled "complicated combinations" of concurrency primitives
that were duplicated across modules into small, business-logic-free,
thoroughly multithreaded-tested utilities in a new common.concurrent
package (and a CoordinationCas helper in the coordination module), then
swap the call sites onto them. Along the way, fix two latent concurrency
defects surfaced by the audit.

New utilities (each with race-injecting concurrent tests):

• BoundedLruCache: thread-safe access-ordered LRU (was an inline
  synchronizedMap(LinkedHashMap{removeEldestEntry}) in BoxEngine).
• TtlCache: read-through TTL cache with loaders run outside the locks
  (was a ConcurrentHashMap<K,(value,expiresAt)> copy in BoxAclStore,
  S3AccessControl, and ClusterRouter).
• KeyedSlidingWindow: per-key bounded sliding window (was a
  ConcurrentHashMap<K,ArrayDeque> guarded by synchronized(deque) in
  MetricsScraper).
• PeriodicGate: run-at-most-once-per-interval throttle via a volatile
  deadline + double-checked lock (was inline in FileCredentialStore).
• CoordinationCas: optimistic upsert/delete-with-bounded-retry over
  CoordinationService (was duplicated in BoxAclStore and CandyboxNode).

Defect fixes (each with a failing-first regression test):

• ClusterRouter opened connections inside ConcurrentHashMap.computeIfAbsent,
  running blocking connect() under a bin lock (a documented anti-pattern).
  Connect outside the map locks now; a lost race closes the surplus.
• BoxEngine checked the idempotency cache only outside the write lock, so
  two concurrent retries of the same token could both apply the mutation.
  Re-check under the lock so the second caller returns the cached result.

All affected module unit suites pass (583 tests, 0 failures); both
regression tests were confirmed to fail when the fix is reverted.

Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01CdFZHjRQrHzD6FgbDgzh1B

[github-actions]

✅ S3 compatibility gate passed

• mode: gate
• result: 192 passed in 87.77s (0:01:27)

[codecov]

Codecov Report

❌ Patch coverage is 84.57447% with 29 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.82%. Comparing base (4bf1f55) to head (c447245).

Files with missing lines	Patch %	Lines
...edatorray/candybox/common/concurrent/TtlCache.java	81.25%	4 Missing and 2 partials ⚠️
.../me/predatorray/candybox/lsm/engine/BoxEngine.java	76.92%	3 Missing and 3 partials ⚠️
...ay/candybox/common/concurrent/BoundedLruCache.java	76.47%	4 Missing ⚠️
...va/me/predatorray/candybox/server/BoxAclStore.java	66.66%	4 Missing ⚠️
...candybox/common/concurrent/KeyedSlidingWindow.java	89.65%	3 Missing ⚠️
...orray/candybox/common/concurrent/PeriodicGate.java	87.50%	1 Missing and 1 partial ⚠️
...atorray/candybox/coordination/CoordinationCas.java	91.30%	1 Missing and 1 partial ⚠️
.../me/predatorray/candybox/client/ClusterRouter.java	94.73%	1 Missing ⚠️
...a/me/predatorray/candybox/server/CandyboxNode.java	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff            @@
##               main      #30   +/-   ##
=========================================
  Coverage     83.82%   83.82%           
+ Complexity      677      674    -3     
=========================================
  Files           162      167    +5     
  Lines          9284     9369   +85     
  Branches       1397     1404    +7     
=========================================
+ Hits           7782     7854   +72     
- Misses          998     1012   +14     
+ Partials        504      503    -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.