[Tests] Fix slow video tensor creation from list of numpy arrays in SmolVLM

[Defalt-Meh]

What does this PR do?

While running SmolVLM tests I noticed this warning in the output:

tests/test_video_processing_common.py:57: UserWarning: Creating a tensor from a list of
numpy.ndarrays is extremely slow. Please consider converting the list to a single
numpy.ndarray with numpy.array() before converting to a tensor.

Tracked it down to prepare_video() in tests/test_video_processing_common.py — it builds a list of numpy arrays (one per frame), then passes that list straight to torch.tensor(). PyTorch really doesn't like this; it ends up iterating over each array one by one instead of doing a bulk conversion.

The fix is simple — consolidate into a single numpy array first, then hand it to torch:

# before
video = torch.tensor(video).permute(0, 3, 1, 2)
 
# after
video = torch.from_numpy(np.array(video)).permute(0, 3, 1, 2)

np.array() was already imported. Output is bit-identical (verified with torch.equal()), warning is gone.

Quick benchmark on Apple M-series, 500 iterations:

Scenario	Before	After	Speedup
What the tests actually use (8 frames, 10x10)	0.37 ms	0.002 ms	~190x
More realistic frames (32 frames, 224x224)	711 ms	0.06 ms	~11,000x

All SmolVLM tests pass (302 passed, 186 skipped, 0 failures). No other instances of this pattern in tests/*.py.

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@yonigozlan @zucchini-nlp