ci-operator namespace should not cache `IMAGE_INSTALLER`

[enxebre]

Today the ci-operator namespace caching is reusing the same installer image on multiple test re-runs.
If it happens that the e2e coverage on the installer does not catch a bug breaking the relation between the installer and one or multiple operators, then any PR in any of those repos will be death lock even if a fix get merged into the installer.
Also the same would apply the other way around. An operator could get CI green running against a cashed outdated version of the installer to realise that is broken or incompatible with latest changes on the installer after it's merged

ci-operator namespace should not cache IMAGE_INSTALLER as this is an open window for bugs to bypass CI validation

[enxebre]

cc @bbguimaraes @petr-muller @smarterclayton @derekwaynecarr

[smarterclayton]

This has always been the design. The chances of this happening is very low in practice.

We should probably make it easier to bust the cache, and possibly explore if there is an unambiguous signal that the cache is truly expired.