(fix): disallow NumpyExtensionArray

[ilan-gold]

This PR disallows NumpyExtensionArray at both the deepest level (PandasExtensionArray) and also at all of the top-level entrance points I could think of that should be auto-fixed. I'll need to add some tests to test that assumption but any help coming up with other potential entrace points would be helpful. For now this PR auto-converts anywhere as_compatible_data is used (which seems to be setting data, initialization, __setitem__ and copy on Variable) as well as auto-converting on dataframe conversion. So any help erroring/auto-converting elsewhere would be great.

I also whitelisted some other data types that I noticed we were touching as well, but I think those may require closer examination so will revert/investigate as needed.

• Closes Regression in DataArrays created from Pandas #10301 (would happily close this if @richard-berg has something more robust/different)
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[dcherian]

Suggested change

	pd.array([1, 2, 3], dtype="int64"),
	pd.array([1, 2, 3], dtype="int64"),
	pd.array([1, 2, 3], dtype="int64[pyarrow]"),

[dcherian]

Can we make this more exhaustive please? with datetime, timedelta etc.

[ilan-gold]

Without opening a can of worms, datetime and timedelta are the two things that are currently (on main) whitelisted (without that, they don't roundtrip).

import pandas as pd

df = pd.DataFrame({"arr": pd.date_range("20130101", periods=4, tz="US/Eastern")})
df.to_xarray()["arr"].to_pandas()

loses the timezone, for example, without the whitelist on from_dataframe.

So I don't think we want to add those here, which is also why I whitelisted them elsewhere. I was hoping that before adding tests, we would settle on what exactly this PR should be doing (of course NumpyExtensionArray but should it cover anything else).

I am happy to roll back that whitelist, i.e., leave it where it was on from_dataframe and then allow these types through anyway via other means into a Dataset.

[dcherian]

I'd simply like a test or two that exhaustively records the current behaviour (whether cast to numpy or not), so we can be sure of what is going on. Two tests (one for preserved dtype, and one for numpy casting) would work fine.

There's also two cases: ExtensionArray in a data variable, and ExtensionArray as an indexed coordinate variable.

[ilan-gold]

Two tests (one for preserved dtype, and one for numpy casting) would work fine.

Not 100% certain what this meant - could you clarify?

There's also two cases: ExtensionArray in a data variable, and ExtensionArray as an indexed coordinate variable.

I think this covered here. When I check xr.Dataset.from_dataframe(df)["arr"].variable it's an IndexVariable as expected when it's an index on the original object

[ilan-gold]

Also check out ilan-gold#2 for a related cleanup. As is here, the data types already round trip so that PR only cleans up the internals a bit and the tests/behavior are also a bit clearer now (better preserved date types). Other than that, no changes in terms of types or behavior, it seems