Play Framework

fix: fourth audit pass — interrupts, ThreadLocal hygiene, doc/spec sync

P2 fixes:
- Mail.java: switch the back-pressure gate from acquireUninterruptibly()
  to acquire() so Future.cancel(true) and resetExecutor() can unblock
  waiters cleanly. Previously a saturated gate would let shutdown leak
  still-pending sends; now cancellation propagates and the task returns
  false with the interrupt status restored.
- PlayHandler.java: clear CachedBoundActionMethodArgs and
  ControllersEnhancer.currentAction in the request-cleanup block.
  ActionInvoker.initActionContext sets both ThreadLocals; without the
  cleanup, pooled platform threads leaked the previous request's bound
  method args and controller-call stack across requests. VT mode masked
  this because each request got a fresh thread that died after.
- ServletWrapper.java: parity fix — also clear
  ControllersEnhancer.currentAction (CachedBoundActionMethodArgs was
  already cleared here). Both paths now have the same cleanup contract.
- VirtualThreadScheduledExecutor.java: SelfReschedulingFuture transitions
  to terminal state via handle.cancel(false) when the scheduler is shut
  down between runs or RejectedExecutionException is thrown during the
  reschedule submit. Previously the periodic future could remain
  non-terminal forever — isDone() returned false, get() blocked
  indefinitely — diverging from STPE's "shutdown cancels periodic
  tasks" contract.
- openspec/specs/java-25-compilation/spec.md: rewrite to describe
  "Java 25 only" reality. Previous spec required compatibleJavaVersions
  to retain entries 17–23 and described default java.source as "17",
  both contradicting the current code (which accepts only "25" and
  defaults to "25").
- documentation/manual/releases/release1.12.x/releasenotes-1.12.0.textile:
  correct two passages that still claimed Java 17–25 acceptance with
  the current "25 only / pre-25 JDKs rejected at startup" behavior.

P3 fixes:
- resources/module-skel/build.xml: bump build-release from 17 → 25 so
  new-module skeletons match the framework's Java 25 emission.
- GroovyTemplate.java: wrap tagTemplate.internalRender(args) in
  try/finally so TagContext.exitTag() runs even when render throws.
  Without this, the TagContext ThreadLocal stack accumulated stale
  entries on pooled threads when render threw.

Deferred:
- StreamChunkAggregator upload-spooling on the Netty EventLoop thread
  (P2): legitimate concern but requires structural refactor — either
  AsynchronousFileChannel, a dedicated I/O executor with bounded queue,
  or migration to Netty's HttpPostRequestDecoder. Out of scope here;
  worth a separate ticket.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>