test: cover osw.utils.util.parallelize

[LukasGold]

Closes #25

Changes

Adds tests/utils/parallelize_test.py. No production code changes.

Covered:

• results returned in input order, not completion order
• kwargs forwarded to the wrapped function
• empty iterable, and a non-list iterable such as range
• tasks genuinely run concurrently
• return_exceptions=False (default): the first exception propagates
• return_exceptions=True: exceptions land at their own index, neighbours survive, every item is still attempted
• the already-running event loop path, which is what notebooks hit

The return_exceptions cases are a regression guard for the silent-drop fix in #132.

Note on the concurrency test

It uses a threading.Barrier rather than wall-clock timing: the barrier only clears if all items are in flight at once, so serialized execution fails deterministically instead of flaking under CI load.

Observation, not changed here

Under MODE = "asyncio", both flush_at_end and progress_bar are inert. flush_at_end is only read in the MODE == "dask" block, and the asyncio branch calls tqdm.gather unconditionally without checking progress_bar. The tests deliberately assert nothing about either, so they do not lock in dead behaviour. Worth a separate issue if those parameters should work or be removed.

[github-actions]

Release preview

No version bump from the current commits (stays at v2.0.2). Use conventional commit types (feat, fix, ...) to trigger a release.

Changelog preview (truncated)

_{Preview via python-semantic-release and conventional commits.}