Fix: coerce unknown types to O dtype

[brynpickering]

Added a check for Python objects that neither have numpy dtype dtype attributes nor are one of the builtin types so that type promotion succeeds and defaults to object dtype.

This deals with cases where user-defined instances are in the array which would otherwise trip up numpy.result_type.

I did think about placing this in xarray.core.dtypes.preprocess_types but chose to place it closest to where it causes issues (namely, on calling xp.result_type).

• Closes Cannot apply sum to custom object arrays with min_count=1 if n_dims <= sum_dims #9755
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[max-sixty]

do you think we could add a quick comment in the code describing what it's doing, for the less informed of us?

(does anyone know this better?)

[brynpickering]

Sure, done.

Another options for this would be to check all the args passed to the compat.array_api_compat.result_type method and if any of them are not a numpy-compatible object or a builtin type then just return np.object_ directly. Passing the args to np.result_type is unnecessary as any object dtype in the list will automatically lead to an object dtype being returned (as it's the lowest common denominator).