Add three-tier model routing with VRAM management and benchmark suite

- Three-tier routing: light (router answers directly ~3s), medium (qwen3:4b
  + tools ~60s), complex (/think prefix → qwen3:8b + subagents ~140s)
- Router: qwen2.5:1.5b, temp=0, regex pre-classifier + raw-text LLM classify
- VRAMManager: explicit flush/poll/prewarm to prevent Ollama CPU-spill bug
- agent_factory: build_medium_agent and build_complex_agent using deepagents
  (TodoListMiddleware + SubAgentMiddleware with research/memory subagents)
- Fix: split Telegram replies >4000 chars into multiple messages
- Benchmark: 30 questions (easy/medium/hard) — 10/10/10 verified passing
  easy→light, medium→medium, hard→complex with VRAM flush confirmed

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

agent_factory.py

@@ -0,0 +1,54 @@
+from deepagents import create_deep_agent, SubAgent
+
+
+def build_medium_agent(model, agent_tools: list, system_prompt: str):
+    """Medium agent: create_deep_agent with TodoList planning, no subagents."""
+    return create_deep_agent(
+        model=model,
+        tools=agent_tools,
+        system_prompt=system_prompt,
+    )
+
+
+def build_complex_agent(model, agent_tools: list, system_prompt: str):
+    """Complex agent: create_deep_agent with TodoList planning + research/memory subagents."""
+    web_tools = [t for t in agent_tools if getattr(t, "name", "") == "web_search"]
+    memory_tools = [
+        t for t in agent_tools
+        if getattr(t, "name", "") in ("search_memory", "get_all_memories")
+    ]
+
+    research_sub: SubAgent = {
+        "name": "research",
+        "description": (
+            "Runs multiple web searches in parallel and synthesizes findings. "
+            "Use for thorough research tasks requiring several queries."
+        ),
+        "system_prompt": (
+            "You are a research specialist. Search the web thoroughly using multiple queries. "
+            "Cite sources and synthesize information into a clear summary."
+        ),
+        "tools": web_tools,
+        "model": model,
+    }
+
+    memory_sub: SubAgent = {
+        "name": "memory",
+        "description": (
+            "Searches and retrieves all relevant memories about the user comprehensively. "
+            "Use to gather full context from past conversations."
+        ),
+        "system_prompt": (
+            "You are a memory specialist. Search broadly using multiple queries. "
+            "Return all relevant facts and context you find."
+        ),
+        "tools": memory_tools,
+        "model": model,
+    }
+
+    return create_deep_agent(
+        model=model,
+        tools=agent_tools,
+        system_prompt=system_prompt,
+        subagents=[research_sub, memory_sub],
+    )