Adds session/class scope support for models to enable efficient GPU resource management when running tests with more models than available GPUs.

Key changes:

• Add @pytest.mark.model(name, scope="session|class") marker support
• Session-scoped models are pre-launched at startup (default)
• Class-scoped models are launched on-demand when needed
• Add LRU eviction: when GPUs are full, least recently used models are evicted
• Evicted models are queued for re-launch when needed again
• Add Gateway class for unified gateway lifecycle management
• Add worker management API tests (IGW mode)

This enables CI environments with limited GPUs (e.g., 4 GPUs, 6 models):

1. Pre-launch what fits (a, b, c, d)
2. Queue overflow (e, f)
3. On-demand: when test needs 'e', evict LRU and launch 'e'

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments (/tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci) or contact authorized users to do so.
4. After green CI and required approvals, ask Merge Oncalls to merge.