I think this boils down to creating a

public interface DatastoreContext extends Identifiable<DatastoreIdentifier>, SchemaContextProvider;

which really is very similar to current MountPointContext, except it is not burdened with MountPointIdentifier (which is not really needed in the codec paths).

This will also allow us bits and pieces of datastore – for example the ability to record a DataTreeSnapshot, which is needed to parse constructs like:

leaf foo {
    type union {
        // references a data tree instance
        type leafref;
        type string;
    }
}

Since parsers are required to normalize the union, they decide whether it matches the leafref before they fall back to string. In order to do that, they need access to a data tree instance on which to evaluate the leafref expression...