Complete and document .torch.random.py

[giovannivolpe]

Complete and document .torch.random.py:

• Revise .torch.random.py to include dtype and device arguments required by torch.
• Add unit tests.
• Add documentation.

[giovannivolpe]

Some comments that might be relevant (from a close PR #332).

• documentation and unit testing of deeptrack.backend.array_api_compat_ext.torch.random. Also verify carefully that the arguments are dealt with correctly.
• Address the following the points raised in PRs Bm/implement-device-dtype #340 and Bm/implement-math #341:

#340
Implements dtype and device on Feature and config.
These should then be used during array creation as xp.zeros((100, 100), dtype=self.float_dtype, device=self.device)

No automatic dtype or device casting happens. We will have to handle this ourselves in the manner described above.
Current implmentation of torch.random module does not support this syntax. We should make sure it does in a separate PR.

#341
This is a test implementation of a few classes in deeptrack.math. It's to demonstrate how I envision the implementation working.
Note:
array creation should always set dtype and device:
xp.zeros((100, 100), dtype=self.float_dtype, device=self.device)

torch does not automatically create arrays from lists. so we should systematically use stack instead (same functionality for lists of arrays).
I also show how unittests can work. We just need to set BACKEND=... to run the tests again with torch backend.