add cross-build functionality

With this change, if a container build job is tagged with a runner on a
different architecture than requested with CI_REGISTRY_IMAGE_ARCH, the
build will make use of qemu and cross-build the container for the
requested architecture.

A single test of a python container build on my notebook yielded
sobering results: 3min native vs. 22min cross build. That makes a
performance loss of around 700%. Nevertheless it may be preferred to
hosting special runners for extra archs.

One step closer to #4