Implement persistent user long-term memory with query-context injection

[AlbertCorry]

Summary

NLWeb's Memory pre-check (AskAgent/python/core/query_analysis/memory.py) can detect when a user asks to be remembered, but it only emits a transient remember message. There is no durable per-user storage, and stored memories are not injected into later query context.

Expected behavior

• When the Memory pre-check detects a remember request, persist the memory for the user (oauth_id or user_id).
• On later queries, prepend stored memories to prev_queries as [memory] {text}.
• Re-enable the Memory pre-check in handler preparation when memory_enabled is true.
• Expose REST endpoints to list and delete user memories:
  • GET /user/memory?user_id=...
  • DELETE /user/memory/{memory_id}?user_id=...
• Support memory_mock=true for offline/testing without LLM calls.

Context

• docs/nlweb-memory.md describes long-term memory as a hook, not a full implementation.
• memory.py line 43: # this is where we would write to a database
• Memory step is commented out in baseHandler.prepare().

Commit reference

b423f15d9aeaa023ce75993ac9deed2354597043