feat(mcp): add configurable small_model support for OpenAI provider

[Milofax]

Summary

This PR adds support for configuring the small_model parameter in the MCP server's config.yaml file.

Problem

Previously, the small model was hardcoded to either gpt-5-nano or gpt-4.1-mini based on the main model type (see factories.py lines 122-135). This made it impossible to use local LLMs like Ollama which don't have these OpenAI-specific models available.

When using Ollama with Graphiti MCP Server, users would get errors like:

model 'gpt-4.1-mini' not found, try pulling it first

Solution

• Add optional small_model field to LLMConfig in schema.py
• Use configured small_model in factories.py if set, otherwise auto-detect as before
• Bonus fix: Also pass base_url from provider config to CoreLLMConfig (was missing, caused issues with OpenAI-compatible endpoints)

Changes

File	Change
mcp_server/src/config/schema.py	Add small_model: str | None field
mcp_server/src/services/factories.py	Use config.small_model if set + pass base_url
mcp_server/config/*.yaml	Add documented small_model option
mcp_server/README.md	Update Ollama example with small_model

Example config for Ollama

llm:
  provider: "openai"
  model: "qwen3:32b"
  small_model: "qwen3:32b"  # IMPORTANT for local LLMs
  max_tokens: 4096

  providers:
    openai:
      api_key: "ollama"
      api_url: "http://localhost:11434/v1"

Test plan

• Verify existing behavior unchanged when small_model not set (auto-detection)
• Test with Ollama: set small_model to same model as main model
• Verify base_url is correctly passed to OpenAI client

Fixes #1155

🤖 Generated with Claude Code

[danielchalef]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[Milofax]

I have read the CLA Document and I hereby sign the CLA

[Milofax]

recheck