Rudder

So, in #19114, we had missing directive because of a bug.
Now, we have the correct behavior, but only after a full regeneration if the added directive doesn't change actual configuration.

This is because is the added directive is skipped everywhere, the effective configuration doesn't change on any nodes, so the first step of policy generation that looks if it is necessary to compute new generation say "no, it isn't", then we don't have new expected configuration, and it's expected configuration that are used to check if directives are skipped.
That's all because we don't really apply directives to rule, we apply directives to nodes (with a lot of graphe resolution in the middle, involving rules and groups).

Correcting that one would mean that we need to change expected configuration in a case when it's actually not needed (because, well, they don't change) and would lead to poetentially long computation for nothing. I'm not sure about what to do.