Add Qwen3 Coder 480B MLX serving

[mweingartner]

Summary

• add Qwen3-Coder-480B-A35B sidecars for MLX 4-bit and UD-Q4_K_XL GGUF
• add MLX server routing through run-server.py and llama-swap model generation
• extend model download support for Hugging Face snapshots and clean markdown/pre-commit lint issues

Validation

• pre-commit run --all-files
• python3 -m py_compile download_model.py route-config.py run-server.py lm_launcher/.py llama_swap/.py
• python3 route-config.py --validate
• python3 download_model.py --list
• served local Qwen3-Coder-480B MLX snapshot on 127.0.0.1:8001 and verified /v1/chat/completions returned "ready"

Note: route-config validation still reports warnings for planned model artifacts that are not present locally.