Unseeded random sizes in test_image_processing.py make coverage nondeterministic

[aymuos15]

test_crop_center_2d and test_pad_centre_2d in tests/core/test_unit/test_tools/test_image_processing.py choose crop/pad sizes with unseeded random.randint:

def test_crop_center_2d():
    x_shape = random.randint(2, 100)
    ...
    new_x_shape = random.randint(1, x_shape)

def test_pad_centre_2d():
    ...
    new_x_shape = random.randint(x_shape, x_shape * 10)

crop_center_2d / pad_center_2d branch on is_even(pad), so whether the even-branch or odd-branch line executes depends on those random sizes. As a result, cellfinder/core/tools/image_processing.py reports a slightly different set of covered lines on each run, with no code change.

Reproduce:

pytest tests/core/test_unit/test_tools/test_image_processing.py \
  --cov=cellfinder.core.tools.image_processing --cov-report=term-missing -p no:xdist

Running it a few times, the missing lines flip between e.g. 77-78, 84 (11 missed) and 77-78, 86-87 (12 missed). Adding random.seed(0); np.random.seed(0) to the two tests makes the missing lines identical on every run.

[Haris-bin-shakeel]

I was able to reproduce this. Coverage does fluctuate between runs, and it appears to come from the random shape generation rather than the implementation itself.

The root cause seems to be pad_center_2d(): the is_even() branch depends on the randomly generated input shape, so different runs exercise different parity paths. I also noticed a small edge case in test_crop_center_2d() when new_x_shape == x_shape.

A minimal fix would be to seed both random and numpy.random at the start of these two tests only (random.seed(0) and np.random.seed(0)). Since there isn't a project-wide seeding fixture, this keeps the change local and makes the tests deterministic without affecting the rest of the suite.

If that approach sounds reasonable, I'm happy to put together a PR.