Fix #111 and maybe minor fixes WIP

[votroto]

Updating of polynomial variables throughout final_touch meant that polynomial and MOI variables were no longer in sync; this resulted in nonsensical models.

My solution just updates the variables, then takes the sorting approach from _scalar_polynomial in nl_to_polynomial.jl. It is not great, but without more significant changes, some variable-order-preserving hocus-pocus will have to be done anyway.

I took the liberty of suggesting an edit for the / overload, which should now try to avoid creating rational polynomials.

The _to_polynomial for ScalarQuadraticFunction in functions.jl seems redundant after issue #110.

Tests that the variables are preserved should be written, but without more significant changes, that might be tricky. Maybe just a few e2e tests in the meantime? I will write a few for the /, at least.

Fixes #111

[votroto]

After more testing, this does not solve the issue.

[votroto]

Ah, it was wrong in the _scalar_polynomial already too.

[blegat]

The changes look good, thanks, just missing a few tests. Yes, unit tests calling the internal functions is good.

[votroto]

Ok, about ready for review. The tests are so complex we might need tests for the tests, but they were failing before and are passing now, and they do cover a lot of the changes. So... success? Is there a standard way to do randomized/property testing in Julia?

I'd also welcome an explanation on what the type T of the _scalar_polynomial function is, because I can't get anything to compile consistently. That might simplify the tests a tiny bit.

If this is good enough, I could try proposing a refactoring PR, if such a thing would be welcome.

BTW, Is it expected for dynamic polynomials with rational coefficients to be a little unstable?

[blegat]

Need to fix format

[odow]

@blegat are the QCQP tests even run?

[odow]

I've dropped the change to a / b. I'll re-add it in a separate PR.