v0.10.5 release

[shoyer]

We had several regressions in v0.10.4, so I'd like to issue a v0.10.5 release shortly with some fixes.

I also had a couple of highly productive plane rides to/from vacation, so I found time to put together a number of bug-fix PRs. If anyone has time to help review these, it would be greatly appreciated!

Issues to resolve:

• How should Dataset.update() handle conflicting coordinates? How should Dataset.update() handle conflicting coordinates? #2180 (resolved for now)

PRs for regressions in v0.10.4:

• Fix writing to groups with h5netcdf BUG: fix writing to groups with h5netcdf #2181
• Fix DataArray.stack() with non-unique coordinates on pandas 0.23 Fix DataArray.stack() with non-unique coordinates on pandas 0.23 #2168
• Fix dtype=S1 encoding in to_netcdf() Fix dtype=S1 encoding in to_netcdf() #2158

PRs for older bugs:

• Fix DataArray.groupby().reduce() mutating input array Fix DataArray.groupby().reduce() mutating input array #2169
• Bug fixes for Dataset.reduce() and n-dimensional cumsum/cumprod Bug fixes for Dataset.reduce() and n-dimensional cumsum/cumprod #2156
• Validate output dimension sizes with apply_ufunc Validate output dimension sizes with apply_ufunc #2155
• Fix string slice indexing for a length-1 CFTimeIndex Fix string slice indexing for a length-1 CFTimeIndex #2166
• Fix unlimited dims bug fix unlimited dims bug #2154

cc @pydata/xarray @crusaderky

[shoyer]

OK, that took a little longer than I thought!

I plan to merge #2158 and issue the 0.10.5 end of day today unless anyone has objections.

[shoyer]

I tagged v0.10.5 and uploaded it to pypi before noticing that due to a bug in our versioneer setup, our version number was 'v0.10.5' in xarray.__version__ instead of the correct '0.10.5'. Oops!

Pypi doesn't allow overwriting files, so I just tagged and uploaded v0.10.6 as well.

Hopefully this version builds properly on conda-forge: conda-forge/xarray-feedstock#33