feat(intelligence): complete migration to Rust backend

- Unify all intelligence modules to use intelligenceClient
- Delete legacy TS implementations (agent-memory, reflection-engine, heartbeat-engine, context-compactor, agent-identity, memory-index)
- Update all consumers to use snake_case backend types
- Remove deprecated llm-integration.test.ts

This eliminates code duplication between frontend and backend, resolves
localStorage limitations, and enables persistent intelligence features.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

desktop/src/store/memoryGraphStore.ts

@@ -6,7 +6,11 @@
 
 import { create } from 'zustand';
 import { persist } from 'zustand/middleware';
-import { getMemoryManager, type MemoryEntry, type MemoryType } from '../lib/agent-memory';
+import {
+  intelligenceClient,
+  type MemoryEntry,
+  type MemoryType,
+} from '../lib/intelligence-client';
 
 export type { MemoryType };
 
@@ -184,8 +188,10 @@ export const useMemoryGraphStore = create<MemoryGraphStore>()(
         set({ isLoading: true, error: null });
 
         try {
-          const mgr = getMemoryManager();
-          const memories = await mgr.getAll(agentId, { limit: 200 });
+          const memories = await intelligenceClient.memory.search({
+            agentId,
+            limit: 200,
+          });
 
           const nodes = memories.map((m, i) => memoryToNode(m, i, memories.length));
           const edges = findRelatedMemories(memories);