fix: race condition for SuspendException

[maschnetwork]

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

Problem

Tests were failing intermittently on GitHub Actions but passing locally. Executions that should
return PENDING (waiting on callbacks/waits) were incorrectly returning FAILED.

Root Cause

The SDK uses SuspendExecutionException to unwind the call stack when execution should suspend.
The DurableExecutor relied on checking suspendFuture.isDone() to detect suspension, but due to a
race condition, the exception could propagate and be caught by generic exception handlers before
the future was visible as done - causing it to be treated as a failure.

This race was more likely to occur on CI runners due to different thread scheduling and slower
execution compared to local machines.

Changes

DurableExecutor.java

• Simplified exception handling to explicitly check for SuspendExecutionException (including when
  wrapped by CompletableFuture)
• Removed redundant isCompletedExceptionally() check - now handled by single catch block

StepOperation.java

• Catch SuspendExecutionException in finally block of worker thread to prevent phaser deadlock
• Without this fix, stepAsync().get() could block forever when the exception escaped the executor

[zhongkechen]

So this is a fix of how we handle suspend exceptions. The PR title/commit message might be confusing as it implies an issue in tests.