Describe the issue:

Hi ,

I have realized that when I pass a masked array in numpy ma diff np.ma.diff and prepend a masked intial value, the mask is deleted and replaced (in my case for 0. values). Despite being the fill_value different from 0.

I would expect that np.ma.diff keeps the mask after calculating the difference, specifically if you are prepending a masked array that keeps the same structure than the input array.
See below:

Reproduce the code example:

import numpy as np
# a is the dummy array
a = np.random.randn(5,10,10)
a = np.ma.masked_where(a<.1,a)
print(a.mask.any()) # returns True

b = np.ma.diff(a, prepend=a[:1,...],axis=0)
print(b.mask.any()) # returns False (unexpected)

# What I would expect / my workaround
c = np.ma.diff(np.ma.append(a[:1], a, axis=0),axis=0)
print(c.mask.any()) # returns True (as expected)

NumPy/Python version information:

My numpy version is: '1.23.4'
Python version is: 3.10.6

Context for the issue:

Shouldn't np.ma module's objects respect always the input mask (if it is properly referenced in its arguments)?

Thanks a lot and keep on with the amazing numpy!