Furtherly optimize the namespace cache by mapping "virtual" namespaces to their real namespace

[ghost]

As a somehow extraneous feature of #2599, I am moving the "virtual" namespace (i.e. real namespace + arbitrary number of "sub-namespaces", dot-separated) resolution out of NamespaceService#resolveWorkspaceIdForNamespace and centralizing it in the in-memory, request-scoped NamespaceRequestCache (which is used by the service and a few more).

This way, the resolution can be used in one point from nearly anywhere in the code base (which also allows removing the same identical logic from the UserNamespaceRoleService - since it uses the cache too).

To develop this further on and optimize on a few CPU cycles / stack entries, the idea would be to map the resolved real namespace by virtual namepace name in a new concurrent map in the cache.
That would require moving the recursive logic in another method that retains the "root" virtual namespace name and queries the map with it / puts it in once resolved.

This is a minor/trivial task but given the volume of resolution requests when dealing with a model in a sub-namespace, it is worth implementing soon.