You can still copy just the primary file and not the WAL or any other files if you want to live dangerously and possibly lose transactions if there are write operations in parallel.
Of course, you'd be better off with WAL-based replication tech like litestream instead of plain file copies if you are truly worried about parallel operations during your file copies.
You can even distribute individual WAL frames and arrange distributed writes with raft consensus and a time lease. I never formally modelled this but it seemed to work perfectly: the approach only lacked checkpointing the Wal file and synchronising the database file across raft nodes.
Of course, you'd be better off with WAL-based replication tech like litestream instead of plain file copies if you are truly worried about parallel operations during your file copies.