Add Ray runtime support

[sercanuste]

This pull request adds support for the Ray Serve LLM runtime, improves GPU fleet navigation, and introduces explicit /dev/shm sizing for LLM deployments to prevent engine startup issues, especially with vLLM and Ray. It also updates the database schema and types to support these features.

Ray Serve LLM runtime support:

• Added a new ray runtime to the LLM deployment options, including configuration, startup command, and special handling for GPU hosts and OpenAI compatibility (src/config/gpu-model-library.json, src/lib/database/provider/types.extended.ts). [1] [2]

GPU fleet navigation improvements:

• Added breadcrumb resolvers for GPU fleet hosts and deployments to improve navigation in the UI (src/components/layout/breadcrumbResolvers.ts).

/dev/shm sizing for LLM deployments:

• Added a shmSizeBytes field to LLM deployment types, database schema, and migration logic, allowing explicit override of shared memory size for containers to avoid vLLM/Ray startup hangs (src/lib/database/provider/types.extended.ts, src/lib/database/sqlite/base.ts, src/lib/database/sqlite/schema.ts). [1] [2] [3]

Development workflow:

• Increased the Node.js memory limit for the dev:watch script to match the main dev script, preventing out-of-memory errors during development (package.json).