Gaussian-blur test is failing

The test at tests/test_transforms.py::test_blur currently fails in a weird way: if run through the full test suite, then a gaussian-blur transform becomes a NOOP - it simply does not work. To reproduce this, you have to run the full suite, and the test is likely to fail in the end.

https://gitlab.idiap.ch/biosignal/software/deepdraw/-/jobs/315366#L1835

I marked the test as xfail for the time being, to allow our migration to continue, but we need to ensure this piece of code is working as intended.

I'm worried that, under certain conditions, gaussian blur data augmentation may not work as intended generating input data copies instead of blurred versions.

@arenzom: can you please check this? Any ideas?