cassini-tests: Fix race condition in K3s integration tests logic

When the test deployment is upgraded to a later container image
version, K3s stops the Pod(s) with the old container image and starts
the Pod(s) with the new container image. The step after upgrading the
deployment is to query the running Pods and check if they are using
the new container image.

This situation represents a race condition. If the query is performed
quickly, then the old Pod hasn't yet been stopped and the old Pod name
is considered to be the new Pod. This causes an error later in the
tests when it does stop, and subsequent queries to it become invalid.

This patch fixes the race condition by ensuring that a full set of
'new' Pods are running after the upgrade, before continuing with the
rest of the test suite.

cassini-docs: Updated to reflect the above mentioned logic.

Changelog: bug

Signed-off-by: Vikas Katariya <vikas.katariya@arm.com>