Enable kdump in VMs and collect dumps from kernel panics

[sharnoff]

ref https://neondb.slack.com/archives/C03TN5G758R/p1690216475962149?thread_ts=1690175203.472599&cid=C03TN5G758R

This should significantly help with debugging in the event of kernel panics in the VMs — which is made much more likely by memory hotplugging.

Tangentially related to #176.