[BUG] `replace` gives incomplete result on non-contiguous view

[pkomiske]

Describe the bug
Calling bottleneck.replace on a non-contiguous view does not replace all occurrences of the old value. This leads to an incorrect result.

To Reproduce
This script reproduces the bug:

import bottleneck as bn
import numpy as np

print('Bottleneck version', bn.__version__)
print('Numpy version', np.__version__)
print()

# make array with some nans
x = np.random.rand(4, 3, 2)
x[::2,:,0] = np.nan

y = x[:,:,0]
print('Before:')
print(y)
print()

# attempt to replace nans
x2 = bn.replace(y, np.nan, 0)
print('After:')
print(x2)

which for me yields

Bottleneck version 1.6.0
Numpy version 2.0.2

Before:
[[       nan        nan        nan]
 [0.6648173  0.68779559 0.22735558]
 [       nan        nan        nan]
 [0.92285548 0.83657247 0.21408579]]

After:
[[0.         0.                nan]
 [0.6648173  0.68779559 0.22735558]
 [0.         0.                nan]
 [0.92285548 0.83657247 0.21408579]]

I would of course expect to see all the nans replaced, but the last one is not.

• Python version is 3.12.6
• OS is Ubuntu 22.04
• pip version is 25.3
• bottleneck and numpy versions are listed above