Fix CRAP issues: Extract JWT utils, workflow builders, polling logic

CRAP Score Improvements:
  unified_synthesis_handler: 52.8 → 22 (57% reduction)
  poll_workflow_result: 38.4 → 0 (REMOVED, split into helpers)

DRY Improvements:
  - Extracted JWT token extraction to handlers/jwt_utils.rs (shared)
  - Extracted workflow builders to handlers/workflow_builder.rs
  - Extracted polling logic to handlers/workflow_poller.rs
  - Removed duplicate code: -50 LOC across modules

Architecture:
  ├─ jwt_utils.rs: extract_jwt_token()
  ├─ workflow_builder.rs: WorkflowBuilder + WorkflowQueryBuilder
  ├─ workflow_poller.rs: poll_workflow_until_complete(), response parsing
  └─ handlers use shared utilities

Testability:
  + 18 new unit tests for builders + polling
  + 6 new unit tests for JWT utils
  + Mock-friendly response parsers (parse_workflow_status, etc.)

SRP Improvements:
  ├─ unified_synthesis_handler: Route + orchestrate (NOT parse/build)
  ├─ execute_reasoning_workflow(): Build + poll + parse (single concern)
  ├─ poll_workflow_until_complete(): ONLY polling (retries, timeout)
  └─ Response parsers: ONLY extraction (no business logic)

Compilation: ✅

docs/TEMPORAL_WORKFLOW_INTEGRATION.md

@@ -115,34 +115,96 @@ VALUES
   ('agent-init-agent1', 'abc123', 'active', NOW());
 ```
 
+## Workflow & Activity Pattern
+
+**Key Insight**: Memory service orchestrates via REST, Temporal activities do the actual LLM work + persistence.
+
+```
+Memory Handler                      Temporal Workflow              LLM API
+    |
+    v
+Extract JWT token
+    |
+    v
+SynthesisClient.execute_workflow(
+  {"action": "START_WORKFLOW", ...}
+)  ─────────────────────────────→  ReasoningWorkflow
+                                      |
+                                      v
+                                  InferenceActivity
+                                      |
+                                      v (LLM call)
+                                   api.riotpiao.com/v1/chat/completions
+                                      |
+                                      v (with JWT)
+                                   LLM Response
+                                      |
+                                      v (persist)
+                                   memory_entity + memory_edge
+                                      |
+                                      v (activity completes)
+                                  Workflow completes
+                                      |
+Poll DESCRIBE_WORKFLOW ←───────────  Return result
+    |
+    v
+Memory handler receives result
+    |
+    v
+Return to client
+```
+
 ## Workflow Lifecycle
 
 ### 1. Register Agent
 ```
 POST /agents
-  → START_WORKFLOW (AgentInitialization)
+  → Extract JWT from Authorization header
+  → SynthesisClient.execute_workflow({"action": "START_WORKFLOW", ...})
+    └─ Gateway: POST /workflow + JWT header
+    └─ Temporal: Start AgentInitialization workflow
+      └─ Activity: Persist to temporal_workflow_links table
   → Store workflow_id, run_id
-  → Return AgentResponse
+  → Return AgentResponse to client
 ```
 
-### 2. Monitor Execution
+### 2. Reasoning Workflow
+```
+POST /memory/synthesis
+  → JWT required (Bearer token)
+  → SynthesisClient.execute_workflow(ReasoningWorkflow)
+    └─ Gateway: POST /workflow + JWT
+    └─ Temporal: Start ReasoningWorkflow
+      └─ Activity: InferenceActivity
+        ├─ Call: POST api.riotpiao.com/v1/chat/completions (with JWT)
+        ├─ Parse LLM response
+        └─ Persist result to memory_entity/memory_edge (temporal_workflow_links tracks link)
+  → Poll DESCRIBE_WORKFLOW (retry up to 30×, 100ms interval)
+  → Wait for status COMPLETED or FAILED
+  → Return ReasoningResult to client
+```
+
+### 3. Durability & Checkpointing
+```
+Activity Crash during LLM call
+  → Temporal replays workflow from last checkpoint
+  → Activity re-executes (idempotent via idempotency_key)
+  → Result persisted to DB (already done from retry)
+  → No data loss
+```
+
+### 4. Monitor Execution
 ```
 GET /agents/{id}
-  → DESCRIBE_WORKFLOW (query Temporal)
-  → Return execution status
+  → Extract JWT
+  → SynthesisClient.execute_workflow({"action": "DESCRIBE_WORKFLOW", ...})
+  → Return execution status (RUNNING|COMPLETED|FAILED)
 ```
 
-### 3. Agent Reasoning Calls
-```
-POST /synthesis/reason
-  → Temporal workflow sends events to agent
-  → Agent uses SynthesisClient.reason_query() to call LLM
-  → Store reasoning traces
-```
-
-### 4. Cleanup
+### 5. Cleanup
 ```
 DELETE /agents/{id}
+  → JWT required
   → CANCEL_WORKFLOW
   → Mark temporal_workflow_links.status = 'archived'
 ```