Cancel silc_server_connect_to_router_retry when connecting.

There exists a crash bug such that an un-cancelled timeout callback for
silc_server_connect_to_retry fires after the connection object has
already been cleaned up.  Any router_retry requests must be cancelled
when we are deleting the associated connect object.  The fix that was
implemented was to cancel silc_server_connect_to_router_retry in
addition to silc_server_connect_to_router when a call to
silc_server_create_connections is made.  (This routine is called when
we are to make new server connections if reconnects are enabled.)

The problem would typically occur after a long enough time with silcd
trying to connect to a router server over and over; there is a race
condition component that can delay the initial use-after-free condition
for some time.