`Dataset.reduce` pass through non-numeric scalars

[mathause]

• [ ] Closes #xxxx
• [ ] Tests added
• [ ] User visible changes (including notable bug fixes) are documented in whats-new.rst
• [ ] New functions/methods are listed in api.rst

Trying to reduce a non-numeric scalar results in an error. This PR passes them through unchanged.

import xarray as xr
ds = xr.Dataset(data_vars={"y": ((), "string")})
ds.mean()

TypeError: the resolved dtypes are not compatible with add.reduce. Resolved (dtype('<U6'), dtype('<U6'), dtype('<U12'))

---

• I did not find an issue for that
• Alternatively we could skip them (this is done with non-numeric arrays)
• Numeric scalars are still reduced

[shoyer]

I'm pretty sure this was intentional -- mean of a string is not well defined.

Could we make this opt-in, by adding a new keyword argument -- perhaps something like ds.mean(skip_nonnumeric=True)?

As you note, there's also the question of whether to drop or pass through non-numeric values. I'm not sure which is more intuitive.

[mathause]

Thanks for the review. I don't have the capacity to implement this, so I close this PR. (I think my scalars were actually coords, so there it was fine.)