Fixes for running jobs on new k8s-provisioned CERN runners

Key changes:

• Set FF_KUBERNETES_HONOR_ENTRYPOINT to force these runners to still use the declared entrypoint for most jobs (so that e.g. GCC8 from SCL is used in SWATCH plugin build jobs), excl. the docker images builds
  • See https://gitlab.docs.cern.ch/docs/Build%20your%20application/CI-CD/Runners/TroubleShooting/docker-entrypoint-used-k8s-runners/ for more info
• Update the kaniko image entrypoint

@carrillo: I tested these fixes in the calo layer-2 repo, using the test2 branch - e.g. see https://gitlab.cern.ch/cms-cactus/projects/calol2/-/pipelines/6241973