polypath asymmetric routing: Handle MSG1 late dupes

If we get a MSG1 and it seems too late for the protocol stage, and it
had the same peer nonce, it was probably an old packet via a different
route.

In theory it might seem like we should do this in SENTMSG3 and 5 too
because we might have had concurrent setup, and higher priority, so the
peer also sent us MSG2 or 4.  But mobile peers have priority.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>