[DomGarguilo]

Closes #4361

Changes

• Fate.shutdown(long, TimeUnit) → Fate.shutdown(Duration). The shutdown wait chain (waitForAllFateExecShutdown, waitForDeadResCleanerShutdown, FateExecutor.waitForShutdown) now passes a single CountDownTimer instead of (start, timeout, unit).
• EventQueue.poll(long, TimeUnit) → EventQueue.poll(Duration). Callers use a static final constant.
• CountDownTimer.startNew(Duration) now caps very large durations (e.g. Duration.ofMillis(Long.MAX_VALUE)) at Long.MAX_VALUE nanos instead of throwing ArithmeticException. This matches the long, TimeUnit overload and Timer.hasElapsed(Duration), which already do this through Timer.toNanos. That helper is now package-private and reused here. Documented in the javadoc and covered by a new test.

Reasoning

• Fate shutdown: Runs once per shutdown, so allocation doesn't matter. The deadline is now one object, and the loops read !countdown.isExpired() instead of repeating System.nanoTime() - start < unit.toNanos(timeout). A raw long start also no longer has to be a nanoTime value by convention.
• EventQueue.poll: Called about every 100ms, but callers pass a static final constant, so there are no new allocations. Gains a stronger type and a named constant.
• CountDownTimer capping: Makes the Duration and long, TimeUnit overloads behave the same, so callers that use Long.MAX_VALUE ms to mean "no timeout" (e.g. TabletServerBatchReaderIterator) can switch to Duration without gaining a crash path.

Co-Authored-By: Claude Opus 5.5

[DomGarguilo]

This is the last batch and will close #4361 once its is merged.