A colleague tried to use EWA resampling on regular (non-dask) numpy arrays that were integers. She got errors about NaN not being able to be converted to an integer so she then specified the fill value with fill=0 and still got the same error. Turns out I have the fill value hardcoded to NaN:

This means integer data isn't actually handled. I need to create some tests for this to prove that there isn't some other reason I hardcoded this.