Refactor repetitions to a separate datastructure

[mstechly]

As @dexter2206 mentioned in PsiQ/bartiq#136 (review):

After reviewing this and the QREF's counterpart of this PR I'm more and more convinced that it would be better if we had a separate class representing "loop" or a "routine with repetitions" (or at least if we had it in Bartiq). This is because a lot of the logic we perform for validation and dispatching (with ifs) could be neatly hidden. For instance, we verify if routine with repetition has one child - but we could have an object that HAS to have one child (which would be validated by Pydantic). (...)

We should evaluate whether we want to introduce such a change in QREF natively or just in Bartiq.