The way peer addresses were exchanged polluted caches with information about
dead nodes. In particular, bootstrapping often took a long time because the
cache of the primary node was mostly useless.

This also fixes bootstrap of registry.

It's normally safe to use --table=0 because most nodes use SLAAC and by default,
any existing default route is deleted.
It's better to abort if someone who explicitely wants to use a separate table
whereas the kernel is limited.

A lot of bug fix

Removing of the bandwidth calculation

When 2 tunnels are estabilshed between the same 2 peers, one of them is deleted

Remove the refresh-rate option

Added new simulation results.

Check for dead tunnels evry 5 seconds instead of waiting for the next tunnel_refresh

in upnpigd : now open a port in tcp too

Changes in the number of connection per peers