feat(mlx): add Apple Silicon video backend

[localai-org-maint-bot]

Description

Adds an Apple-Silicon-only mlx-video backend for LTX-2 and converted Wan2.1/Wan2.2 checkpoints through LocalAI's existing POST /video API.

• maps LocalAI video requests to the pinned upstream MLX-Video CLI without shell interpolation
• supports text-to-video and image-to-video, plus LTX-2 end-frame conditioning
• wires Darwin backend builds, backend-gallery capability selection, and backend capability discovery
• documents installation and model configuration, with a runnable example

Closes #8085

Notes for reviewers

The pinned upstream currently exposes generation as CLI-oriented functions that load their pipeline within a generation invocation. This backend keeps that lifecycle rather than relying on upstream internals that do not provide a stable reusable pipeline API.

Wan checkpoints must already be converted to MLX and available as a local directory. The upstream LTX-2 path can resolve supported Hugging Face repositories.

Test plan

• python3 backend/python/mlx-video/test.py (4 tests pass)
• python3 -m py_compile backend/python/mlx-video/backend.py backend/python/mlx-video/mlx_video_utils.py backend/python/mlx-video/test.py
• bash -n backend/python/mlx-video/install.sh backend/python/mlx-video/run.sh backend/python/mlx-video/test.sh
• YAML parse for backend/index.yaml, .github/backend-matrix.yml, and the example config
• make test-ci-scripts (41 tests pass)
• git diff --check
• go test ./core/config/... reached 499/500 specs; the only failure is the existing network-dependent config validation test because this environment returns HTTP 403 for raw.githubusercontent.com

A native backend package/model generation smoke test requires an Apple Silicon runner and is left to Darwin CI.