Auto detect MTGPU

[yeahdongcn]

Testing Done

❯ make build
CGO_ENABLED=1 go build -ldflags="-s -w" -o model-runner ./main.go
❯ ./model-runner
WARN[0000] Could not read VRAM size: could not get nvidia VRAM size 
INFO[0000] Running on system with 31859 MB RAM          
INFO[0000] Successfully initialized store                component=model-manager
INFO[0000] LLAMA_SERVER_PATH: /Applications/Docker.app/Contents/Resources/model-runner/bin 
INFO[0000] Supported MTHREADS GPU detected, MUSA will be used automatically 
INFO[0000] Metrics endpoint enabled at /metrics         
INFO[0000] Supported MTHREADS GPU detected, using MUSA variant