Always set status to RollingUpdate if operation exists

[victor-carvalho]

When trying to do operation that requires to recreate the pods (change image, env vars, etc ...) and there are no pods the operation gets stuck until it timeouts. This happens because on GetOperationStatus it checks if the operation is in RollingUpdate status and if not the progress always returns zero and EnsureCorrectRooms do not set the status if the number of invalid pods is zero.

This PR makes EnsureCorrectRooms always set the status to RollingUpdate if there is a current operation even if there are zero pods to update.

[coveralls]

Coverage decreased (-0.03%) to 74.533% when pulling 851e2ff on fix/update-scheduler-config-zero-replicas into 82080ea on master.

[gabrielcorado]

Can this change mess with some Operation that has finished?

EDIT: As we discussed it will be a problem only if two workers get it, but since Operation key is deleted at API when it finishes we think that is an edge case that is very unlikely to happen. Btw, I suggested a log addition so we can track what the description was before we change it to rolling update.

[leohahn]

Is it hard to add a regression test for this issue?