feat(chat): add automatic memory extraction mode

[basnijholt]

Summary

The chat agent currently uses a tool-based approach for memory - the LLM must explicitly call add_memory and search_memory tools. This gives users control (LLM asks permission before storing), but misses the automatic fact extraction that memory-proxy provides.

Current Behavior

• Memory tools are available to the LLM
• LLM decides when to store (and asks user permission per system prompt)
• No automatic memory retrieval injected into context
• User doesn't see when memories are stored/retrieved

Proposed Enhancement

Add an --auto-memory flag (or similar) that enables the full memory-proxy pipeline:

1. Auto-retrieve: Inject relevant memories into LLM context before each turn
2. Auto-extract: Extract and store facts from each conversation turn
3. Visual feedback: Show user what memories were retrieved/stored

Possible Modes

Mode	Retrieval	Storage	Use Case
--no-memory	None	None	Privacy, testing
--memory (current)	Tool-based	Tool-based	User control
--auto-memory	Automatic	Automatic	Seamless experience
--auto-memory-retrieve	Automatic	Tool-based	Hybrid (best of both?)

Implementation Notes

• Could leverage memory_client.chat() which already implements the full pipeline
• Or call augment_chat_request() + extract_and_store_facts_and_summaries() directly
• Need to consider: should auto-mode still respect the "ask permission" guideline?

Related

• PR feat(chat): integrate vector-backed memory system with LLM tools #183 - Current tool-based memory integration
• agent_cli/memory/engine.py - process_chat_request() has the full pipeline

[basnijholt]

Implemented as part of PR #183 in commit 05edfdd. The --memory-mode auto option now enables automatic fact extraction after each conversation turn.