feat(openai): add inference GPT Live model

[russellmartin-livekit]

Summary

• add InferenceGPTLiveModel and its native session transport for LiveKit Inference
• sign LiveKit inference auth and connect to /v1/live/sessions?model=openai/gpt-live-1
• preserve direct OpenAI GPT-Live support for all provider tiers while narrowing LiveKit Inference to the metered default tier
• fail unsupported inference tiers before connecting and document that hosted OpenAI tools are unavailable through inference
• classify gateway startup errors using the completed session.started future rather than assuming a non-empty session id
• preserve the native GPT-Live payload and direct OpenAI transport behavior

Testing

• uv run pytest tests/test_gpt_live_model.py tests/test_realtime/test_openai_inference_gpt_live_model.py — 76 passed
• focused OpenAI mypy
• Ruff check and formatting

Rollout

Merge after the outstanding GPT-Live correctness fixes settle and after agent-gateway#1399 reaches staging.