Fix "concurrent update in progress" errors during canary deploys

[rianmcguirefly]

Change Summary

What and Why:

Avoid possibility of failure during a canary deploy if the machine that we run the canary update on is currently stopped.

Example error:
Failed to update machines: failed to update machine <machine id>: failed to update VM <machine id>: aborted: machine is replacing: concurrent update in progress Retrying...

The machines API blocks concurrent updates to a machine, so we need to ensure that doesn't happen.

How:

Fix all the things

• When updating a non-started machine, we now wait for the machine update to fully complete (eg. machine transitions from replacing back to stopped) before continuing
• Avoid updating the canary machine twice
• Prefer a started machine for the canary where possible

[dangra]

@rianmcguirefly I merged this after it passed all tests so it is included in today's auto release. thanks