Deterministically ordered `BinaryPolynomial`

**Application**
The terms of a `BinaryPolynomial` are implemented with `frozenset`.
Python sets are unordered in a non-deterministic way. This leads to some interesting behavior that propagates to the related functions, `reduce_binary_polynomial` and `make_quadratic_cqm`.

Within a python invocation, it is possible to produce the same CQM model repeatedly, given the same `BinaryPolynomial` object.
This changes from run to run though, see this [post](https://stackoverflow.com/questions/3848091/set-iteration-order-varies-from-run-to-run#:~:text=The%20reason%20the%20set%20iteration,because%20of%20the%20random%20seed) on stackoverflow. This can lead to CQM models with different variables and constraints given the same `BinaryPolynomial`, or different reduced terms and constraints being returned from `reduce_binary_polynomial`.

It would be useful to be able to produce the same reduction results from run to run, rather than having to fix an environment variable to force the behavior.

**Proposed Solution**
Rework `BinaryPolynomial` and related functions to avoid sets - or possibly use something like an [ordered set](https://github.com/rspeer/ordered-set) instead.

**Alternatives Considered**
Setting the environment variable `PYTHONHASHSEED` makes set iteration deterministic. This has to be done prior to invoking Python.