Fix: Propagate terminal worker failures and stop queued DAG work

[mday-io]

Summary

Worker cancellation could leave callers waiting indefinitely. Propagate the original exception and stop queued work after a terminal failure.

- unfinished future or queued work after failure
+ propagate failure, stop queued work, drain active workers

• Preserve independent work when collecting ordinary errors.
• Prerequisite for Feat: Add optional per-task context propagation #6141 and Feat: Add optional execution observations #6051.

Test Plan

• Reproduced the original hangs and queued-work failure.
• Tested both error modes, worker draining, exception identity and executor reuse.
• Ran make style and all make fast-test batches: 2,862 passed.
• Confirm current-head CI, including Windows.

[nicksspirit]

Took a look. Catching BaseException so a Ctrl-C or SystemExit in a worker actually reaches the caller instead of hanging it, plus having queued nodes bail once fail-fast trips, makes sense. Looks good.