interp can not handle datetime64, while fix seems easy

The interp function can not handle datetime64 dtypes (docs are not explicit about this; they state that x and xp should be values). Error message is that array cannot be cast to dtype('float64') according to the rule 'safe'. But conversion seems easy.

Reproducing code example:

import numpy as np
d1 = np.datetime64('2019-01-01')
d2 = np.datetime64('2019-01-03')
xp = np.array([d1, d2])
yp = np.array([1, 3])
x = np.datetime64('2019-01-02')
np.interp(x, xp, yp)  # produces error

while the following works fine

np.interp(x.astype('float64'), xp.astype('float64'), yp)

So why is that not done in the interp function?

Start with the np.interp entry point and reproduce the datetime64 failure using the example in this issue. Check how datetime64 arrays are currently cast and compare that behavior with the working float64 conversion. Done means the documented datetime64 example runs successfully with the expected interpolated value, with coverage for the relevant input types.