Refactor create_llm in agent_langchain.py to reduce cyclomatic complexity

[barry-hollywood]

Refactor create_llm() in langchain_agent/agent_langchain.py to reduce cognitive complexity

This is in order to reduce SonarQube findings on the maintainability of the code base ( issue #212 ).

Key Changes

I've refactored create_llm() by offloading the logical but complex-to-read if/elif chain into separate provider factory functions that can easily self-contain that logic while making the original function easier to read and understand.

The refactoring splits the monolithic if/elif chain (checking provider types) into five dedicated helper functions:

• _create_openai_llm()
• _create_azure_llm()
• _create_bedrock_llm()
• _create_ollama_llm()
• _create_anthropic_llm()

Benefits

The key benefits of these changes are to reduce cognitive complexity of the create_llm() function.

Separating the logic like this means that we can more easily make changes to specific provider implementations without having to navigate through large logical flows that handle multiple different providers. Each provider' configuration and validation logic is now isolated and independently testable.

Files Changed

• agent_runtime/langchain_agent/agent_langchain.py - main refactoring
• agent_runtime/langchain_agent/tests/test_agent_langchain.py - unit tests for all provider functions

Testing Notes

Comprehensive tests have been added covering:

• All five LLM providers
• Required credential validation
• Optional parameter handling
• Missing dependency scenarios
• Common args construction

Note: Local test suite has pre-existing import errors with langchain on both main and this branch, unrelated to these changes. The refactored code passes syntax validation and only modifies the create_llm function structure without touching imports or dependencies.