For the VM's themselves, ignore that they are virtual machines. Treat them like a physical host, and install snmp. Monitor them in the normal fashion then.
As for #1. that is more of a problem to do. You could likely script the Xen server to event when a host goes down, and deal with the results. Zenoss responds quickly when data is passed to it, but is slower when it has to go and hunt down changes due to the periodic way it does modeling.
As for the alerts in email for the VM hosts, once they are monitored define a group called VM or something like that for them. Then simply make an email alert rule stating if the alarm is for a host in group VM and criticality is >= (whatever) send the email alert..