Make illegal path-like variable names when constructing a DataTree from a Dataset

[etienneschalk]

• Closes When creating a DataTree from a Dataset with path-like variable, subgroups are expected to be created #311
• Tests added
• Passes pre-commit run --all-files
  - [ ] New functions/methods are listed in api.rst
• Changes are summarized in docs/source/whats-new.rst

Technical Note

Regarding Hashable vs str Dataset keys

Note: DataTree keys are Hashable. I only check for slashes in the variable names if they are instance of str.
I never encountered a case (yet) where a Dataset keys are not str but Hashable in the broader case.
We can imagine corner-cases where keys would be other types of Hashable, eg Path from pathlib

In [2]: from pathlib import Path

In [3]: hash(Path("/"))
Out[3]: -3809984204556177651

The choice I made is (1): only apply the check of slashes in the key if the key is an instance of str.
Another choice (2)would be to project the Hashable space onto str space: str(variable_name)
(1) seems more conservative than (2) as I do not pretend to be able to get a string representation for any Hashable.

[TomNicholas]

@etienneschalk sorry for missing this - do you want to re-submit this PR to xarray upstream?

[TomNicholas]

Closing in favour of all the upstream versions of this issue - see links above.