refactor: split api.py monolith into 19 route modules (Phase B)

Split 3,598-line api.py into thin orchestrator (128 lines) + 19 route
modules in bot/routes/:

  core.py (7 routes), mood.py (10), language.py (3), evil_mode.py (6),
  bipolar_mode.py (9), gpu.py (2), bot_actions.py (4), autonomous.py (13),
  profile_picture.py (26), manual_send.py (3), servers.py (6),
  figurines.py (5), dms.py (18), image_generation.py (4), chat.py (1),
  config.py (7), logging_config.py (9), voice.py (3), memory.py (10)

All 146 routes verified present via test_route_split.py (149 tests).
21/21 regression tests (test_config_state.py) pass.
Monolith backup: bot/api_monolith_backup.py (revert: cp it to api.py).

bot/routes/gpu.py

@@ -0,0 +1,38 @@
+"""GPU selection routes."""
+
+from fastapi import APIRouter, Request
+from utils.logger import get_logger
+
+logger = get_logger('api')
+
+router = APIRouter()
+
+
+@router.get("/gpu-status")
+def get_gpu_status():
+    """Get current GPU selection"""
+    from config_manager import config_manager
+    return {"gpu": config_manager.get_gpu()}
+
+
+@router.post("/gpu-select")
+async def select_gpu(request: Request):
+    """Select which GPU to use for inference"""
+    data = await request.json()
+    gpu = data.get("gpu", "nvidia").lower()
+
+    if gpu not in ["nvidia", "amd"]:
+        return {"status": "error", "message": "Invalid GPU selection. Must be 'nvidia' or 'amd'"}
+
+    try:
+        from config_manager import config_manager
+        success = config_manager.set_gpu(gpu)
+
+        if success:
+            logger.info(f"GPU Selection: Switched to {gpu.upper()} GPU")
+            return {"status": "ok", "message": f"Switched to {gpu.upper()} GPU", "gpu": gpu}
+        else:
+            return {"status": "error", "message": "Failed to save GPU state"}
+    except Exception as e:
+        logger.error(f"GPU Selection Error: {e}")
+        return {"status": "error", "message": str(e)}