fix(memory): infinite loop with auto_ingest mode.

[acivan]

This pull request introduces a new example for using Memori with a custom OpenAI client and improves the handling of internal agent calls by switching from pattern-based detection in message content to a more reliable metadata flag approach. It also consistently tags internal memory processing calls with a specific metadata flag across memory and retrieval agents.

Key changes:

New Example:

• Added examples/supported_llms/openai_custom_example.py, which demonstrates how to integrate Memori with a custom OpenAI client, including environment variable setup, memory tracking, and interactive chat.

Internal Call Detection Improvements:

• Updated _is_internal_agent_call in memori/integrations/openai_integration.py to check for internal processing using the metadata field instead of searching for patterns in message content, reducing false positives and improving reliability.

Consistent Internal Metadata Tagging:

• Added the "INTERNAL_MEMORY_PROCESSING" metadata tag to all internal memory processing calls in both MemoryAgent (memori/agents/memory_agent.py) and RetrievalAgent (memori/agents/retrieval_agent.py), ensuring these calls are properly identified as internal. [1] [2] [3] [4]

[harshalmore31]

@acivan Thanks for the contribution, this is more robust architectural solution to the infinite loop problem, if possible can you adapt the tests cases similar to the #115 to verify the metadata-based approach in PR, This would give us both the proper fix AND thorough test coverage