Add optional config knob for cpu overcommit limiting

[tpepper]

We observe notable slowness when actively booting qemu vCPU's are 40x overcommitted to physical CPU's. One of our demos is to launch a tonne of workloads. The OS in the workload is very busy booting for a brief spell at start up, but then quiets down. This is a normal startup pattern. Rather than trying to start 10k vm's in parallel, things run better if you launch in parallel until things are "too busy", then launch more as the first ones are becoming less busy.

This knob would tie to the "some multiple" mentioned in issue #99.