Fix error reporting in Schedulers - #2036

Original PR: AVSystem#15

Summary

In JDK 25 ForkJoinPool implements ScheduledExecutorService. That change exposed a long-standing gap in Scheduler
error reporting - the failure of a scheduled task could be lost.

Problem

An uncaught failure of a task given to scheduleOnce, scheduleAtFixedRate or scheduleWithFixedDelay did not reach
the Scheduler's UncaughtExceptionReporter when the scheduler was backed by a bare ScheduledExecutorService:
the executor captures the failure in the Future it returns, and Monix used to discard that Future.

Before JDK 25 this stayed mostly invisible, because Monix's own builders either avoided that path or supplied
an AdaptedThreadPoolExecutor, which reports such failures through its own afterExecute hook.
From JDK 25 Scheduler.computation and Scheduler.forkJoin are backed by a ForkJoinPool that takes that path,
so they lost the failures of their scheduled tasks too.

On Scala.js the analogous gap was in AsyncScheduler.scheduleOnce, which handed the task to setTimeout without the
reporting wrapper that execute uses.

Changes

1. The scheduled methods wrap the task so its failure reaches the scheduler's UncaughtExceptionReporter. Failures that NonFatal does not match are reported and then re-thrown. This covers Scheduler.computation and Scheduler.forkJoin on JDK 25.
2. The Scala.js AsyncScheduler.scheduleOnce gets the wrapping that its execute already uses, so a failure that NonFatal matches reaches the scheduler's UncaughtExceptionReporter instead of the ExecutionContext. Its periodic methods are built on scheduleOnce, so they report their failures too.
3. Stopping after failure (the behaviour scheduleAtFixedRate already documented) is now documented on scheduleWithFixedDelay as well (matching their behavior and JDK docs).
4. Reordered InterceptRunnable.apply so a null reporter short-circuits before the TrampolinedRunnable case, returning the runnable unwrapped instead of a wrapper that NullPointerExceptions on the first failure.

Not addressed here

• On a Scheduler over a caller-supplied, plain ExecutorService, the fatal failures of scheduled tasks still reach the thread uncaught exception handler instead of the reporter.
• On a Scheduler over a caller-supplied ScheduledExecutorService, a fatal failure of a task given to execute, or to scheduleOnce with a non-positive delay, is still invisible (Scheduler.computation and Scheduler.forkJoin on JDK 25 are unaffected).
• A reporter installed with withUncaughtExceptionReporter is still ignored by the scheduled methods on
  • the wrappers returned from withUncaughtExceptionReporter and withExecutionModel
  • on Scheduler.singleThread and Scheduler.fixedPool
• Scheduler.singleThread and Scheduler.fixedPool still ignore the reporter they are given when reportFailure is called from outside their own threads.
• Fatal failures on ExecutionContext-backed schedulers still reach the ExecutionContext instead of the reporter.

Incidental cleanup

• Dropped the unused deprecated auxiliary constructors of the private ExecutorScheduler subclasses FromSimpleExecutor and FromScheduledExecutor. A matching exclusion is added to MimaFilters.
• ComputationSchedulerSuite builds its scheduler with Scheduler.computation rather than Scheduler.forkJoin, which ForkJoinSchedulerSuite already covers.