Don't node_load original node on new revisions

In filefield_field_update, the code checks if it's processing a new node, and if it is, it returns early as there can't be any old items to delete. The very same check could be used on new node revisions, as the files won't be deleted either and it saves a few cycles not to node_load the original node.

The performance hit isn't huge in normal operations, but bites quite more when batch processing a lot of nodes :-)