Revert "Fix: MPS allocator error in model.to() transfer (#305)"

This reverts commit 9424020687.

src/optimization/memory_manager.py

@@ -875,31 +875,7 @@ def _standard_model_movement(model: torch.nn.Module, current_device: torch.devic
         debug.start_timer(timer_name)
     
     # Move model and clear gradients
-    try:
-        model.to(target_device)
-    except RuntimeError as e:
-        # MPS allocator workaround: Some PyTorch/macOS versions fail with bulk transfers
-        error_msg = str(e).lower()
-        is_mps_alloc_error = (
-            torch.device(target_device).type == "mps" and 
-            any(keyword in error_msg for keyword in ["watermark", "allocat", "memory"])
-        )
-        
-        if is_mps_alloc_error:
-            if debug:
-                debug.log("MPS bulk transfer failed, moving parameters individually", 
-                        category="memory", indent_level=1)
-            
-            # Move parameters and buffers individually to avoid allocator bug
-            with torch.no_grad():
-                for param in model.parameters():
-                    param.data = param.data.to(target_device)
-                for buffer in model.buffers():
-                    buffer.data = buffer.data.to(target_device)
-        else:
-            # Re-raise if it's a different error
-            raise
-
+    model.to(target_device)
     model.zero_grad(set_to_none=True)
     
     # Clear VAE memory buffers when moving to CPU