xarray issue with reduction on not all present expected_groups

[LunarLanding]

The test below fails before this commit and succeeds after.
However getting the correct fill_value, if any, and avoiding computation so the current tests do not fail will require further work.

def test_fix_xarray():
    import flox.xarray
    import xarray as xr
    import numpy as np

    sa=10
    sb=13
    sc=3

    x=xr.Dataset({
        'v0':xr.DataArray(((np.arange(sa*sb*sc)/sa)%1).reshape((sa,sb,sc)),dims=('a','b','c')),
        'v1':xr.DataArray((np.arange(sa*sb)%3).reshape(sa,sb),dims=('a','b')),
    })

    r=flox.xarray.xarray_reduce(x['v0'],
                              x['v1'],x['v0'],
                              expected_groups=(np.arange(6),np.linspace(0,1,num=5)),
                              isbin=[False,True],
                              func='count',
                              dim='b',
                              fill_value=np.nan,
                             ).mean('a',skipna=True)
    assert len(r['v1'])==6
test_fix_xarray()

[dcherian]

Thanks @LunarLanding for the test and attempting to fix it. I have a simpler fix in #112. Can you try that and see if it looks like expected?

[LunarLanding]

Oh my, it is much simpler, thanks! I found an issue when I started testing the dask case, fixed it like this LunarLanding@10996a8 , tests run ok.

[dcherian]

Oh nice catch. Can you open a pull request to that branch? That way, you can get some credit too