Fix: Use correct generator in __iter__ method Ensure the local gener… (…

…#4078)

Fix: Use correct generator in __iter__ method  Ensure the local generator is used in the random number generation process to maintain reproducibility.

## Summary
This commit fixes an issue in the `__iter__` method where the wrong generator was being used in the random number generation process. The local `generator` variable, which is correctly initialized based on the condition, should be used instead of `self.generator`.

## Changes
- Updated the `torch.randint` call to use the local `generator` variable.
- Ensured that the random number generation process is consistent and reproducible.

## Impact
This change ensures that the random number generation process is correctly controlled by the local `generator`, which is crucial for maintaining the reproducibility of experiments and results.

## Testing
- Added unit tests to verify that the `__iter__` method generates the expected indices.
- Ran existing tests to ensure no regressions were introduced.

src/otx/algo/samplers/balanced_sampler.py

@@ -113,7 +113,7 @@ def __iter__(self):
                 index = torch.cat(
                     [
                         self.img_indices[cls_indices][
-                            torch.randint(0, len(self.img_indices[cls_indices]), (1,), generator=self.generator)
+                            torch.randint(0, len(self.img_indices[cls_indices]), (1,), generator=generator)
                         ]
                         for cls_indices in self.img_indices
                     ],