fix(mlx-vlm): install torch dependencies on Metal (#11164)

Some Transformers processors used by MLX-VLM, including Qwen vision models, import both PyTorch and Torchvision. Include them in the Metal backend environment so model loading does not fail with missing-library errors.

Assisted-by: Codex:gpt-5

Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>

Author: localai-org-maint-bot

backend/python/mlx-vlm/requirements-mps.txt

@@ -1 +1,3 @@
-git+https://github.com/Blaizzy/mlx-vlm@v0.4.4
+git+https://github.com/Blaizzy/mlx-vlm@v0.4.4
+torch
+torchvision