Unification laws test

[ClarkeRemy]

This branch adds a large test for unification, putting Prolog and MORK head to head to check MORK unification correctness.

With a check for cycles added, they agree on the results.

The test is done by running main in experiments/unification_test_laws.

A little bit of the complexity comes from the fact that, to have the test finish in a reasonable amount of time, they run fully in parallel; the Rust and Prolog code. On my machine it takes about 10 minutes per language to finish unifying. I've done my best to avoid over-optimizing however, as this isn't meant to be a benchmark.

This uses SWI Prolog, so one needs to install SWI Prolog to run the test.

[Adam-Vandervorst]

Looks good except the unifiable change; ExprZipper should be seen as deprecated: the unifiable test should instead inline the unify method but use the coroutine version of apply with an Io::sink.

[ClarkeRemy]

I made the change in the unification test, I get the same results. Which is good.

I push the change, but I do feel that doing this inlining reveals that this pattern of inlining is done in a lot of parts of the code.
How many code paths will need to be changed to check if there were cycles that weren't before? I see 11 other calls for apply_e.

I think this pull request can stop here, and we can address the occurs check on the cycles btreemaps at those 11 call sites in another pull request.