fix(Raster): reclassify np.float64 correctly

[martclanor]

Just a small fix on dtype reclassification when creating a Raster object.

I was also tempted to reclassify float as FLOAT64 and int as INT64 because based on my quick search, Python's float is represented as 64-bit, and int is pretty much unbounded. But then maybe these might be intended for optimization purposes as FLOAT32 and INT32 are practically sufficient in most cases.