[udp] use ICMP error messages to trigger faster link down detection

this solves a possible race condition when:

• node1 is running
• node2 very fast
• node1 does NOT have enough time to detect that node2 has gone and reset the local seq numbers / buffers
• node1 will start rejecting valid packets from node2

There is still a potential minor race condition where app
can restart so fast that kernel / network don't have time
to generate an ICMP error. This will be addressed using
instance id in onwire v2 protocol, as suggested by Jan F.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>