allow create Ray cluster with reduced dependency

[dongwang218]

Why ?

Start Ray cluster without the extra vllm dependency

How ?

Add ray as a optional dependency.

so now we have

• client: pip install -e .
• with ray: pip install -e .[ray] "ray[default,serve]==2.52.0"
• with vllm: pip install -e .[vllm_0112]

Test plan

uv pip install -e .[ray] "ray[default,serve]==2.52.0"

matrix start_cluster --add_workers 1 --slurm '{"account": data, "qos": h100_lowest}'

matrix deploy_applications --applications '[{"name": hello, "app_type": "hello"}]'

matrix check_health --app_name hello


matrix deploy_applications --action add --applications "[{'name': 'container2', 'app_type': container, 'min_replica': 1, 'max_ongoing_requests':16, 'ray_resources': {'num_cpus': 1, 'num_gpus': 0.5}}]"

matrix check_health --app_name container2
output:     CUDA available: True Device count: 1 tensor([[0.3865, 2.2107, 0.4473],         [0.4179, 0.1789, 1.0019]], device='cuda:0')
returncode: 0

of course deploy vllm app will fail