instantiate devices from migration payload

[pfmooney]

During a migration, the current logic requires the target instance to be initialized with devices exactly matching the source. Those devices are then "rehydrated" with the migration payloads exported from the source, when that phase of the migration occurs. After discussions with @gjcolombo and others, one recurring conclusion was that it'd be nice if we could instantiate the devices directly from the migration payload.

Doing so would require some restructuring, particularly around what resources are required by each respective device when it is created. Viona, for example, requires the underlying vnic resource for initialization today. Moving it to a model similar to block devices, where it is attached to a "backend" resource prior to being pressed into service, might make sense.

An initialization pattern like this may help free us from demanding that the source and target VM instances adhere to similar internal structures when it comes to device naming, etc.

[hawkw]

This would make some of the work I started a bit ago on IRQ pins a lot nicer!