Fuzz testing

[NMAC427]

It would be greate if at some point we could implement automatic fuzz testing to ensure that there are no bugs hiding. There is an incredible amount ways in which a user can combine expressions and verbs. Using manual testing, we'll never be able to cover all possible cases.

There are many different things that we could test with fuzzing, but probably the simplest would be to randomly generate valid symbolic expressions and then running them on different backends to check that they all either produce the same result or raise NotImplementedError.

(Randomly generating valid expressions, while not trivial, should be doable, because for each operator we have a type signature).