Fixes Numpy datatype compatibility issues

[saurabhkgp21]

Fixes #8207
Taking input as float16 datatype will give segmentation fault, since there are no corresponding types for it in C language. Here I am taking input as float64 type to avoid datatype incompatibility.

[pv]

This is not the right fix. The check for float16 needs to be done in the C code of ndimage.

[saurabhkgp21]

@pv I will update this PR with float16 check in C code

[saurabhkgp21]

@pv please review and see if changes are required or not.

[ghost]

IMO adding a test for this is trivial:

from scipy.ndimage.filters import gaussian_filter
import numpy as np
gaussian_filter(np.array([1],dtype=np.float16), 1)

[saurabhkgp21]

Okay, I will add test cases for this one

[saurabhkgp21]

@pv @xoviat I had added corresponding test cases, please check if everything is fine or not.

[ghost]

Nit: indent with four spaces instead of a tab.

[saurabhkgp21]

ping @xoviat

[saurabhkgp21]

@xoviat is this PR ready to be merged or some more work is needed?

[pv]

I rebased this and replaced the magic number check array_type > 12 with an explicit list of types.

[pv]

thanks, merged