feat(models): add OCI Generative AI provider for Google Gemini on OCI

Adds first-class support for Google Gemini models hosted on Oracle Cloud
Infrastructure (OCI) Generative AI service — a native Google × OCI model
partnership that makes Gemini available directly through OCI's inference
endpoints.

Key design points:
- Subclasses BaseLlm following the anthropic_llm.py pattern
- Uses the OCI Python SDK directly (no LangChain dependency)
- Optional dependency: pip install google-adk[oci]
- Supports API_KEY, INSTANCE_PRINCIPAL, and RESOURCE_PRINCIPAL auth
- Both non-streaming (_call_oci) and streaming (_call_oci_stream) paths
  share setup code via _build_chat_details(); streaming collects OCI's
  OpenAI-compatible SSE events in a thread pool (asyncio.to_thread) and
  yields partial then final LlmResponse
- Registers google.gemini-* (and other OCI-hosted) model patterns in
  LLMRegistry via optional try/except in models/__init__.py
- 37 unit tests (fully mocked, no OCI account needed)
- 10 integration tests (skipped when OCI_COMPARTMENT_ID is unset)

Supported models: google.gemini-*, google.gemma-*, meta.llama-*,
  mistralai.*, xai.grok-*, nvidia.*

Author: fede-kamel

pyproject.toml

@@ -176,6 +176,8 @@ toolbox = ["toolbox-adk>=1.0.0, <2.0.0"]
 
 slack = ["slack-bolt>=1.22.0"]
 
+oci = ["oci>=2.126.0"]                          # For OCI Generative AI model support
+
 [tool.pyink]
 # Format py files following Google style-guide
 line-length = 80

src/google/adk/models/__init__.py

@@ -63,3 +63,13 @@
 except Exception:
   # Gemma3Ollama requires LiteLLM: pip install google-adk[extensions]
   pass
+
+# Optionally register OCIGenAILlm if oci package is installed
+try:
+  from .oci_genai_llm import OCIGenAILlm
+
+  LLMRegistry.register(OCIGenAILlm)
+  __all__.append('OCIGenAILlm')
+except Exception:
+  # OCI support requires: pip install google-adk[oci]
+  pass