Fix BinningAutoBatcher/InFlightAutoBatcher ignoring model.memory_scales_with

When constructing autobatchers from a user-supplied dict, propagate
model.memory_scales_with as the default so fixed-neighbor models like
FairChem use the correct scaling metric instead of the constructor default.

Author: orionarcher

src/atomate2/torchsim/core.py

@@ -244,6 +244,7 @@ def process_in_flight_autobatcher_dict(
             state, model, autobatcher=autobatcher_dict, max_iterations=max_iterations
         )
     else:
+        autobatcher_dict.setdefault("memory_scales_with", model.memory_scales_with)
         autobatcher = InFlightAutoBatcher(model=model, **autobatcher_dict)
 
     autobatcher_details = _get_autobatcher_details(autobatcher)
@@ -284,6 +285,7 @@ def process_binning_autobatcher_dict(
         # pop max_iterations if present
         autobatcher_dict = deepcopy(autobatcher_dict)
         autobatcher_dict.pop("max_iterations", None)
+        autobatcher_dict.setdefault("memory_scales_with", model.memory_scales_with)
         autobatcher = BinningAutoBatcher(model=model, **autobatcher_dict)
 
     autobatcher_details = _get_autobatcher_details(autobatcher)