Reprojection changes floating point precision

[NimSed]

Hi there.

This very line:

reproject/reproject/interpolation/core.py

Line 95 in 3bb79cd

array = np.asarray(array, dtype=float)

changes an input array with dtype=float32 to float64.

Could we replace it with a mere check (for all floating point types) instead, and raise an error in case the input isn't a float?

[NimSed]

I just created this PR: #354
Please feel free to scrap it if it doesn't make sense.

[NimSed]

Update: Of course, the PR fails for many of the dtypes in this test 🤕
But how did it pass the test before, since ALL_DTYPES seems to contain, all dtypes, including i, u (here) 🤔