dtype for GDAL CInt16, rasterio complex_int16

[scottyhq]

• Closes TypeError: data type 'complex_int16' not understood #352
• Tests added
• Fully documented, including docs/history.rst for all changes and docs/rioxarray.rst for new API

Will follow up with a simple test and docs...

[scottyhq]

@snowman2 would it be OK to put a small (100x100 Cint16) test here test/test_data/input/cint16.tif ? or is your preference for creating test datasets for example https://github.com/mapbox/rasterio/blob/master/tests/test_complex_dtypes.py#L37

[snowman2]

@scottyhq thanks for the fix 👍. Either way works for me. If you do add the raster, please compress it.

[snowman2]

@scottyhq thanks for you patience with this. It is looking good. I added a couple of comments for things that would be good to consider.

[scottyhq]

@snowman2 I added more tests to cover dealing with masking and nodata. Given the single dtype= kwarg in to_raster which by default matches the in-memory np.dtype, and the fact that CInt16 is a special case, I think it's easiest to just require an explicit dtype='complex_int16' if you want to write to that format. So I added a note in the docstring.

Tests pass for me locally, except for test_integration_merge.py, which seems unrelated to this PR, and is due to rounding changes in rasterio>1.2.5: (rasterio/rasterio#2202)

[snowman2]

@scottyhq this looks great, thanks 👍

[snowman2]

@scottyhq any issues if I squash and merge?

[scottyhq]

@scottyhq any issues if I squash and merge?

Fine by me! Thanks for the guidance @snowman2