Run Narwhals tests on a single worker and fix issues

[scott-routledge2]

Changes included in this PR

Some tests like those in unique_test.py have non-deterministic results that are both correct, to avoid too many testing fixes, we can mark these types of tests as "skip if multiple workers". We can also catch different issues running tests on one worker vs multiple. Narwhals tests are pretty quick to run (~3 minutes to run both single and multiple works)

• Fixes slicing issue with negative step that returns an empty slice if stop is unbounded and start falls within len(md_head)

Testing strategy

Extended test_slice to add a case for negative step

User facing changes

Bug with slicing a Series with negative step fixed

Checklist

• Pipelines passed before requesting review. To run CI you must include [run CI] in your commit message.
• I am familiar with the Contributing Guide
• I have installed + ran pre-commit hooks.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 68.81%. Comparing base (c33fbb5) to head (1659cd5).
⚠️ Report is 120 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #923      +/-   ##
==========================================
+ Coverage   66.68%   68.81%   +2.13%     
==========================================
  Files         186      195       +9     
  Lines       66795    67571     +776     
  Branches     9507     9594      +87     
==========================================
+ Hits        44543    46502    +1959     
+ Misses      19572    18241    -1331     
- Partials     2680     2828     +148     

[DrTodd13]

If loc also goes through this same new routine then create a test for it as well?

[DrTodd13]

Will .loc also go through this routine? If so, then is stop = None correct? Should start here also get a similar negative check that stop has?

[scott-routledge2]

Since the stop is exclusive, there is no way to represent the beginning of a list in reverse order since 0 would exclude the first element and -1 would imply the last element, which is why I am using None here. Since the start is inclusive then I don't think there needs to be a similar negative check.

From testing I think loc takes this same code path but it seems to work. Any particular reason it wouldn't work here?

[scott-routledge2]

In the tests, we cover missing/None indices, positive indices, and negative indices, the part that was missing was negative step, which I add in this PR, so we should have all the cases covered.

[ehsantn]

Thanks @scott-routledge2 .