implement a ConstrainedPolynomialModel class

**Application**
This would allow inexperienced user to specify optimization problems at a high a level, without needing to worry about technical concerns such as making objectives and constraints quadratic' and to build up models from understandable components.

**Proposed Solution**
Introduce a ConstrainedPolynomialModel class that subsumes all current model types and features

- higher order terms
- all variable types (Binary, Spin, Discrete, Integer)
- (higher order) constraints
- arithmetic operations

The polynomial reduction algorithm can be extended to other variable types. Higher order constraints can be reduced the same way as objective functions. Constrained models can be combined arithmetically by taking the union of constraints of the arguments, making CPMs closed under addition, subtraction multiplication and non-negative integer powers.