[Datasets] Handle all Ray errors in task compute strategy.

[clarkzinzow]

If a Ray error other than a RayTaskError is raised

1. while waiting for map tasks to finish, then we won't properly cancel other map tasks,
2. while waiting for map tasks to be cancelled, then we won't raise the correct (original) task failure exception.

This PR fixes this by catching RayError in both cases, which is a superclass of RayTaskError, TaskCancelledError, and WorkerCrashedError (which may be raised during graceful task cancellation due to a Ray Core bug). This should deflake the test_select_columns test.

Related issue number

Closes #30644

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[c21]

Is there a way to add a unit test for this?

[clarkzinzow]

Confirmed that this deflakes the test.

@c21 I'm attempting to add a test that parametrizes test_dataset_retry_exceptions such that it crashes the worker (sys.exit() or os._exit(0)) to directly test this, but the existing test keeps segfaulting on actor destruction locally... I've been running into this issue with actor garbage collection when running tests locally for a while.

ray/python/ray/data/tests/test_dataset.py

Line 5118 in 8f3e22d

def test_dataset_retry_exceptions(ray_start_regular, local_path):

[clarkzinzow]

Test failures are unrelated, merging.