studio: fix cancel to actually kill llama-server during loading

The unload endpoint checked is_loaded (requires healthy=True), but
during initial loading the server is not yet healthy. Cancel had no
effect because the unload route fell through to the Unsloth backend.

Fix: add is_active property (process exists, loading or loaded) and
check it in the unload route so cancel kills llama-server even during
the download/loading phase.

Also: toast cancel button now properly triggers the backend unload.

studio/backend/core/inference/llama_cpp.py

@@ -58,6 +58,11 @@ class LlamaCppBackend:
     def is_loaded(self) -> bool:
         return self._process is not None and self._healthy
 
+    @property
+    def is_active(self) -> bool:
+        """True if a llama-server process exists (loading or loaded)."""
+        return self._process is not None
+
     @property
     def base_url(self) -> str:
         return f"http://127.0.0.1:{self._port}"

studio/backend/routes/inference.py

@@ -343,11 +343,11 @@ async def unload_model(
     Routes to the correct backend (llama-server for GGUF, Unsloth otherwise).
     """
     try:
-        # Check if the GGUF backend has this model loaded
+        # Check if the GGUF backend has this model loaded or is loading it
         llama_backend = get_llama_cpp_backend()
-        if (
-            llama_backend.is_loaded
-            and llama_backend.model_identifier == request.model_path
+        if llama_backend.is_active and (
+            llama_backend.model_identifier == request.model_path
+            or not llama_backend.is_loaded
         ):
             llama_backend.unload_model()
             logger.info(f"Unloaded GGUF model: {request.model_path}")