vLLM fallback

Use a local vLLM OpenAI-compatible server as an Arka fallback provider.

Start vLLM, then enable it in Arka’s fallback chain:

export VLLM_HOST=127.0.0.1:8000
export VLLM_MODEL=your-loaded-model
export VLLM_FALLBACK=1
arka doctor

Arka tries vLLM when it is reachable or explicitly configured, then continues through the normal provider chain if the local server fails. Set VLLM_FALLBACK=0 to disable the opt-in fallback behavior.