Add global random seed for deterministic tests

[allrob23]

Description

This PR introduces a global random seed set after imports in the test file. This change aims to make the tests more deterministic and easier to debug by ensuring consistent random behavior across test runs.

Current Issues (Before this PR):

The tests currently use random without setting a fixed seed, which causes several problems :

a. Flaky Tests: Tests sometimes pass and sometimes fail due to different random values;
b. Hard to Reproduce: When a test fails, it's difficult to reproduce the exact conditions;
c. Inconsistent CI/CD: Different CI runs may have different outcomes;
d. Time Wasted: Developers spend time debugging issues that are hard to replicate.

(Im not saying these problems are happening, but they can happen.)

Solution

Set a global random seed right after imports in the test file.

Benefits

a. Reproducibility: All test runs will use the same random values;
b. Easier Debugging: When a test fails, we can easily reproduce the issue;
c. Consistent CI/CD: Every CI run will have the same behavior;
d. Clear Intent: It's immediately obvious that we're using controlled randomness;
e. Time Saved: Less time spent on debugging non-deterministic test failures.

[t-bz]

I support this change. As torch.rand() is used in one of the tests, I think we should either fix both seeds or limit the code to a single generator. The same change should also be applied to the keras model tests.

[allrob23]

@t-bz Setting the seed for torch.rand is a great idea. I’ve updated the PR to include torch.manual_seed(42) in all tests that use PyTorch. Additionally, I’ve applied a similar approach to the tests for Keras models

[pluflou]

Thanks for your contributions @allrob23. Could you rebase and fix the conflicts? There's been some changes to the codebase. Thanks!