feat: same-turn memory retrieval in provider mode

prefetch() now reuses the in-flight background search when it matches
the current query, or falls back to a synchronous search (capped at 3s,
split across both requests in case-insensitive mode) so memories are
injected on the same turn instead of one turn behind.

- Track _prefetch_query to scope cached results to their query
- Reset _prefetch_result when starting a new background search to
  prevent stale cross-turn injection after a timed-out join
- Guard background thread writes against superseded queries
- Add optional per-request timeout override to LocalMem0Client
- Update AGENTS.md and README to describe the same-turn flow

README.md

@@ -5,7 +5,7 @@ Self-hosted Mem0 memory provider for Hermes-Agent. Provides semantic memory sear
 ## Features
 
 - **Local Mem0 server** — No cloud dependency, full data privacy
-- **Async prefetch** — Memory retrieval happens in background (~40ms)
+- **Same-turn prefetch** — Memory retrieved synchronously before the LLM call (~40ms)
 - **Context injection** — Relevant memories injected directly into LLM prompt
 - **Automatic fact extraction** — Server-side LLM extracts facts from conversations
 - **Semantic search** — Find memories by meaning, not keywords
@@ -138,16 +138,15 @@ hermes gateway restart
 
 ### How It Works
 
-1. **User message received** → `queue_prefetch()` spawns background thread
-2. **Mem0 search** → Semantic search for relevant memories (~40ms)
-3. **Context injection** → Results injected via `pre_llm_call` hook
-4. **LLM receives** → User message + memory context (no tool call needed!)
+1. **User message received** → `prefetch()` searches Mem0 synchronously (~40ms)
+2. **Context injection** → Results injected before the LLM call
+3. **LLM receives** → User message + memory context (no tool call needed!)
 
 **Example**:
 ```
 User: "Hey, is a new episode out from my favorite anime?"
 
-↓ [Background: mem0.prefetch() searches for "favorite anime"]
+↓ [mem0.prefetch() searches for "favorite anime"]
 
 LLM receives:
 """