v1.2.1-kata-ci

Bugfix:

• cpusoaker should run all cases in each runtimeclass sequentially rather than interleaved. The problem with interleaving them is that the Prometheus metrics gathered during a run may partially reflect the previous run, as Prometheus gathers data over a 30 second interval, which may result in incorrect per-pod memory calculation if one runtimeclass consumes much more memory per pod than another.