feat: Migrate from llamastack to OGX 0.5.2

[gurvinder]

Migrate from llamastack to OGX 0.5.2

The llamastack/distribution-remote-vllm Docker image was archived as part of the Llama Stack → OGX rename. This PR migrates the entire deployment to use ogxai/distribution-starter:0.5.2.

Changes

Helm Charts

• Updated llama-stack image from llamastack/distribution-remote-vllm:0.2.12 to ogxai/distribution-starter:0.5.2
• Rewrote ConfigMap to use the 0.5.2 config schema (centralized storage section, registered_resources, new provider config format, removed deprecated telemetry API)
• Replaced --yaml-config CLI args with RUN_CONFIG_PATH and CLI_NAME=llama env vars (new entrypoint format)
• Added VLLM_API_TOKEN env var passthrough for LiteLLM-proxied endpoints
• Fixed port mismatch: react-agent now connects to http://llama-stack:80 (matching the K8s service port)

React Agent (Streamlit UI)

• Upgraded base image from ubi9/python-311 to ubi9/python-312 (required by llama-stack-client>=0.5.0)
• Pinned client SDK to llama-stack-client>=0.5.0,<0.6.0 (0.6.0+ removes the agents API)
• Updated model listing to use OpenAI-compatible model.id (replaces model.identifier / model.model_type)
• Updated MCP tool format from string ("mcp::hr-api-tools") to dict ({"type": "mcp", "server_label": ..., "server_url": ...})
• Removed unsupported response_format and sampling_params kwargs from ReActAgent
• Fixed streaming event parsing to use 0.5.x structure (chunk.event.delta.text / chunk.event.final_text)
• Display only the thought field from the agent's JSON response

Other

• Updated react-agent image repository to quay.io/rh-ai-quickstart/llama-stack-react
• Filters out safety/guard models when selecting the inference model