Drain worker nodes before upgrading MCR

[kke]

So it would respect Kube PodDisruptionBudgets (+ configurable drain timeout).

Right now it could cause downtime even to applications that follow best practices and have multiple replicas spread on multiple workers

re: @dockerpac

[james-nesbitt]

This issue is old, but has been brought up again. I had to close the old PR because it was exceptionally out of date - but I will look to gather requirements here again.

[james-nesbitt]

The current scope of this effort can use some further discussion

[james-nesbitt]

Their may be a conflict between this and #530