Support non-str Hashables in DataArray

[headtr1ck]

• Closes DataArray types must be strings #8546
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst

Probably we should add a whole bunch of tests for this. For now only testing the constructor.

[TomNicholas]

Probably we should add a whole bunch of tests for this. For now only testing the constructor.

I think I've said this before but eventually I would like to see us use property-based testing and test using a strategy that generates arbitrary strings / hashables. We could build off of #6903 and generalise the names strategy that was introduced in #8404, for example.

[max-sixty]

Probably we should add a whole bunch of tests for this.

We could add them onto a canonical test object like #8533 (maybe there's a param for "object with lots of quirks"), and then if we can spread that object throughout the test suite, we get every method tested on the full range of quirks

---

(I see @TomNicholas just replied with a suggestion to use hypothesis — agree — I think that would provide more depth (more quirks!), though less wide coverage, given we're unlikely to write every test using hypothesis)

[headtr1ck]

Should we explicitly disallow using None as dimension/variable names?
It will inadvertently lead to problems later on.