Take care of overload name when replacing aten op with custom op in to_edge_transform_and_lower (#4010)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/4010

Today when we're replacing an ATen op with the custom op in `to_edge_transform_and_lower` we're not taking care of the overload name. This was working fine till now because most of the ops we're dealing with had the default overload. huiyujie ran into an issue where he was dealing with `torch.ops.aten.upsample_nearest2d.vec` with the vec overload and it was causing issues for him.

Reviewed By: mcr229

Differential Revision: D58710825

fbshipit-source-id: fdbe0c603e68e6db88d38595fa1a10edb512701e