Ensure that _FillValue and statistical attributes have the same data type as the correspondind layer

[gshiroma]

This PR ensures that _FillValue and statistical attributes have the same data type as the corresponding layer.

[bhawkins]

This PR ensures that _FillValue and statistical attributes have the same data type as the corresponding layer.

That makes sense for _FillValue. Do you always want that for stats, though? If you compute the mean of an integer-valued image in floating point it might be preferable to report it that way.

[gshiroma]

This PR ensures that _FillValue and statistical attributes have the same data type as the corresponding layer.

That makes sense for _FillValue. Do you always want that for stats, though? If you compute the mean of an integer-valued image in floating point it might be preferable to report it that way.

Thank you for your comment, @bhawkins. We do not compute statistics for the mask layer. Also, note that this code is implemented in _get_attribute_dict(), where the default value for the to_data_format_function parameter is lambda x: x. This means that, by default, the data type of _FillValue and the statistical values is not modified. This function is (indirectly) called by save_dataset(), which is only used by GCOV and by the code that geocodes RSLC LUTs. So, I don’t see this concern occurring in the current products. But it's a good point. Even if that happened, there could be cases in which it could be preferable to compute statistics of integer datasets in float or integer.

For context, I added this feature after noticing that some statistical values were being saved as float64 for float32 datasets, and this PR fixes that problem.

[bhawkins]

Makes sense.

[hfattahi]

LGTM.

[hfattahi]

@gshiroma let's merge this unless if you are waiting for something else?