Improve how `sh:and` is handled

[bartkl]

Given a property shape p which refers to a node shape N ("superclass"). Furthermore, say we have node shapes N1, N2, ... Nm ("subclasses") which all refer to N using sh:and. Then, we want the type of the Avro record field (corresponding to p) to be union(N, N1, N2, ..., Nm) (and including null if the cardinality requires it).

Note that discussion is ongoing on whether N itself should be included in the record field type. For now we've chosen to do so, but later this might be subject to change.

[bartkl]

Here too, a reasoner might be needed for best practices.

[bartkl]

Correction: the resulting record field type ought not to be the union of all Nm, but the conjugation, which is probably a new record type with the fields of all Nm aggragated there.