want mechanism for tracking guest vCPU pause time

One important piece of data toward tackling #324 is measuring guest vCPU pause time during a migration. In #347 we added some basic instrumentation for measuring migration protocol times on a single host, which gives some data about how long the migrating guest is paused. But the guest pause time must necessarily must be measured across both hosts; we can measure how long the protocol takes, but there may be substantial work on the dest side after protocol completion before the dest can resume vCPUs (such as repairing Crucible volumes).

One option floated by @gjcolombo is having the source transmit performance statistics as a part of the protocol, such that the dest can collect them and export them.

This seems viable, but I also think we will have consider failure cases carefully in such an approach -- e.g., if the migration fails before that data gets transmitted, we still want to have an accurate view of the downtime of the guest.

Start by reading issues #324 and #347, especially the migration-protocol instrumentation added in #347. Trace how source and destination coordinate vCPU pause and resume across migration, including destination work such as repairing Crucible volumes. Done means guest pause time is accurately tracked across both hosts, including migration failures before statistics are transmitted.