Implement prefilter in dask_image.ndinterp.map_coordinates using the dask_image spline_filter function

[GenevieveBuckley]

Currently, if you run map_coordinates with the keyword argument prefilter=True, the results do not exactly match the equivalent results from scipy. See this docstring comment by Marvin:

dask-image/dask_image/ndinterp/_map_coordinates.py

Lines 186 to 190 in 9992f3f

	Warning: prefilter is True by default in
	`scipy.ndimage.map_coordinates`. Prefiltering here is performed on a
	chunk-by-chunk basis, which may lead to different results than
	`scipy.ndimage.map_coordinates` in case of chunked input arrays
	and order > 1.

As a hacky workaround, currently we have prefilter=False as the default in dask-image, but this is not ideal behaviour. It's not great because that does not match the default behaviour of the scipy.ndimage.map_coordinates function.

@astrofrog suggests

What if dask_image.ndinterp.map_coordinates did this automatically, e.g. using dask_image.ndinterp.spline_filter automatically if prefilter=True?
I think once this release is out, it would be worth investigating properly implementing prefilter=True using dask-image's spline_filter as you noted, so that prefilter could actually default to True here.

I think that's a great idea. Then we would get identical results from the scipy and dask-image map_coordinates functions when prefilter=True.

Rough approach:

1. In the dask_image.ndinterp.map_coordinates function, add an initial step - if the prefilter kwarg is True, then add a step manually prefiltering the array using the dask_image.ndinterp.spline_filter function.
2. Then change the default kwargs
   a. Set prefilter to False in the output = da.map_blocks(_map_single_coordinates_array_chunk, ... prefilter=False) part (since the input arrays are already pre-filtered with the dask-image spline interpolation
   

   dask-image/dask_image/ndinterp/_map_coordinates.py

   Line 238 in 9992f3f

   prefilter=prefilter,

   
   b. Set prefilter to True in the outer dask-image map_coordinates function signature (in order to match the scipy map_coordinates implementation)
3. Add a test for this, ensuring the results from scipy's map_coordinates match the dask-image map_coordinates results when prefilter=True. Probably you could extend the existing test with pytest input parameters, instead of having to write an entirely new test.