Add parallel tool calling support and comprehensive integration tests

[fede-kamel]

Summary

This PR adds parallel tool calling support for Meta Llama 4+ models and comprehensive integration tests to ensure quality and compatibility.

Changes:

• Parallel tool calling support for Meta Llama 4+ models via bind_tools(tools, parallel_tool_calls=True)
• Validation that prevents parallel tool calls on unsupported models (Llama 3.x)
• 4 new integration test files with 97 tests total
• LangChain 1.x compatibility verified
• Backwards compatibility with LangChain 0.3.x verified

Test Evidence

Test Count Summary

Test Suite	Count
Unit Tests	47
Integration Tests	97
Total	144

New Integration Test Files

1. test_langchain_compatibility.py (18 tests)

   • Basic invoke, streaming, async
   • Tool calling (single, multiple, parallel)
   • Structured output (function calling, JSON mode)
   • Response format tests
   • LangChain 1.x specific tests

2. test_chat_features.py (20 tests)

   • LCEL chain tests (simple, with history, batch)
   • Async chain invocation
   • Streaming through chains
   • Advanced tool calling with execution
   • Structured output (complex schemas, extraction)
   • Model configuration (temperature, max_tokens, stop)
   • Error handling
   • Conversation patterns

3. test_multi_model.py (30+ tests)

   • Meta Llama models (4-scout, 4-maverick, 3.3-70b, 3.1-70b)
   • Cohere models (command-a, command-r-plus)
   • xAI Grok models (grok-3-70b, grok-3-mini-8b)
   • OpenAI models (gpt-oss-20b, gpt-oss-120b)

4. test_tool_calling.py (modified)

   • Fixed import for LangChain 1.x compatibility

LangChain Compatibility Testing

LangChain 1.1.0 (Latest)

langchain-core==1.1.0
langchain==1.1.0

• Unit tests: 47/47 passed (100%)
• Integration tests: All core functionality verified with real OCI inference

LangChain 0.3.x (Backwards Compatibility)

langchain-core==0.3.80
langchain==0.3.27

• Unit tests: 47/47 passed (100%)
• Full backwards compatibility confirmed

Models Tested (Real OCI Inference - Chicago Region)

Meta Llama

Model	Basic	Streaming	Tool Calling	Structured Output
meta.llama-4-scout-17b-16e-instruct	✅	✅	✅	✅
meta.llama-4-maverick-17b-128e-instruct-fp8	✅	✅	✅	✅
meta.llama-3.3-70b-instruct	✅	✅	✅	✅
meta.llama-3.1-70b-instruct	✅	✅	✅	✅

Cohere

Model	Basic	Streaming	Tool Calling	Structured Output
cohere.command-a-03-2025	✅	✅	✅	✅
cohere.command-r-plus-08-2024	✅	✅	✅	✅

xAI Grok

Model	Basic	Streaming	Tool Calling	Structured Output
xai.grok-3-70b	✅	✅	✅	✅
xai.grok-3-mini-8b	✅	✅	✅	✅

OpenAI

Model	Basic	Streaming	Tool Calling	Structured Output
openai.gpt-oss-20b	✅	✅	✅	✅
openai.gpt-oss-120b	✅	✅	✅	✅

Parallel Tool Calling Feature

Only enabled via bind_tools() parameter (consistent with other LangChain vendors):

from langchain_oci import ChatOCIGenAI

llm = ChatOCIGenAI(
    model_id="meta.llama-4-maverick-17b-128e-instruct-fp8",
    service_endpoint="https://inference.generativeai.us-chicago-1.oci.oraclecloud.com",
    compartment_id="MY_COMPARTMENT_ID",
)

# Enable parallel tool calling (Llama 4+ only)
llm_with_tools = llm.bind_tools(
    [get_weather, calculate_tip, get_population],
    parallel_tool_calls=True
)

Validation ensures:

• Llama 3.x models raise ValueError if parallel_tool_calls=True
• Only Llama 4+ models can use parallel tool calls
• Clear error message guides users to supported models

Test Plan

• Unit tests pass (47/47)
• Integration tests with real OCI inference
• LangChain 1.x compatibility verified
• LangChain 0.3.x backwards compatibility verified
• Multi-model vendor testing (Meta, Cohere, xAI, OpenAI)
• Parallel tool calling validation tests

[YouNeedCryDear]

Duplicate of #59 ?