use np.expand_dims instead of np.newaxis

[rcaneill]

The use of [np.newaxis, lon] is the distance function in geostrophy.py was not compatible with xarray (see exemple below). This PR fixes this issue.

Exemple

import xarray as xr
import gsw
da = xr.DataArray([0,1])
gsw.distance(da, da)

outputs

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-59-24e6827f537c> in <module>
----> 1 gsw.distance(da, da)

~/.cache/pypoetry/virtualenvs/gsw-xarray-NsrEXKiZ-py3.8/lib/python3.8/site-packages/gsw/_utilities.py in wrapper(*args, **kw)
     60             kw['p'] = newargs.pop()
     61 
---> 62         ret = f(*newargs, **kw)
     63 
     64         if isinstance(ret, tuple):

~/.cache/pypoetry/virtualenvs/gsw-xarray-NsrEXKiZ-py3.8/lib/python3.8/site-packages/gsw/geostrophy.py in distance(lon, lat, p, axis)
    185     if lon.ndim == 1:
    186         one_d = True
--> 187         lon = lon[np.newaxis, :]
    188         lat = lat[np.newaxis, :]
    189         axis = -1

~/.cache/pypoetry/virtualenvs/gsw-xarray-NsrEXKiZ-py3.8/lib/python3.8/site-packages/xarray/core/dataarray.py in __getitem__(self, key)
    739         else:
    740             # xarray-style array indexing
--> 741             return self.isel(indexers=self._item_key_to_dict(key))
    742 
    743     def __setitem__(self, key: Any, value: Any) -> None:

~/.cache/pypoetry/virtualenvs/gsw-xarray-NsrEXKiZ-py3.8/lib/python3.8/site-packages/xarray/core/dataarray.py in _item_key_to_dict(self, key)
    703         if utils.is_dict_like(key):
    704             return key
--> 705         key = indexing.expanded_indexer(key, self.ndim)
    706         return dict(zip(self.dims, key))
    707 

~/.cache/pypoetry/virtualenvs/gsw-xarray-NsrEXKiZ-py3.8/lib/python3.8/site-packages/xarray/core/indexing.py in expanded_indexer(key, ndim)
     48             new_key.append(k)
     49     if len(new_key) > ndim:
---> 50         raise IndexError("too many indices")
     51     new_key.extend((ndim - len(new_key)) * [slice(None)])
     52     return tuple(new_key)

IndexError: too many indices

[ocefpaf]

Looks like there is a regression in latest pip on Windows. Getting out of build isolation did the trick for now.

@efiring this PR LGTM. Do you mind taking a look and merging if you agree?

[efiring]

Minor suggestions; if you disagree, it can go in as-is.