fix: PREFER_AMD_GPU now lives in globals so config API changes affect GPU routing

Previously gpu_router.py had its own module-level PREFER_AMD_GPU constant
that was frozen at import time. The config API wrote to globals.PREFER_AMD_GPU
which didn't exist, so runtime GPU preference changes never took effect.

Now globals.py owns PREFER_AMD_GPU and gpu_router reads it from there.

bot/globals.py

@@ -26,6 +26,7 @@ VISION_MODEL = os.getenv("VISION_MODEL", "vision")
 EVIL_TEXT_MODEL = os.getenv("EVIL_TEXT_MODEL", "darkidol")  # Uncensored model for evil mode
 JAPANESE_TEXT_MODEL = os.getenv("JAPANESE_TEXT_MODEL", "swallow")  # Llama 3.1 Swallow model for Japanese
 OWNER_USER_ID = int(os.getenv("OWNER_USER_ID", "209381657369772032"))  # Bot owner's Discord user ID for reports
+PREFER_AMD_GPU = os.getenv("PREFER_AMD_GPU", "false").lower() == "true"  # Runtime-overridable via config API
 
 # Cheshire Cat AI integration (Phase 3)
 CHESHIRE_CAT_URL = os.getenv("CHESHIRE_CAT_URL", "http://cheshire-cat:80")