Do a commit for the output file when flushing for Copy.

I think the output file flush should commit the data to stable storage
in case the server reboots during the copy_file_range call.
This way, only the Copy RPC needs to be redone for this situation.