Allow custom/extensible memory_type values beyond episodic/semantic/message

Problem

The MemoryTypeEnum is hardcoded to ["episodic", "semantic", "message"] and rejects any other values. In practice, agents need additional memory types like pattern, decision, gotcha, tip to categorize knowledge more precisely.

Current behavior: creating a memory with memory_type="pattern" raises a validation error.

Proposal

Either:

1. Make memory_type a free-form string with the current values as documented conventions (not enforced enum)
2. Extend the enum to include commonly useful types: pattern, decision, gotcha, tip, procedure
3. Allow configuration of additional memory types via server config

Option 1 is simplest and most flexible — agents are better at categorizing when not constrained to a small enum.

Context

The MCP tool description already documents semantic vs episodic well. Agents naturally want to tag memories with richer types ([gotcha], [pattern], [decision]) that aid retrieval. The validation blocks this without adding safety value — memory_type is metadata, not a control field.

Files

agent_memory_server/long_term_memory.py — MemoryTypeEnum class

Start in agent_memory_server/long_term_memory.py at the MemoryTypeEnum class and trace where memory_type validation is applied when creating a memory. Compare the proposed free-form, extended-enum, and server-configuration approaches before choosing a direction. Done means the accepted memory types match the decided behavior and the current documented conventions remain supported.