Implement composition polynomials

[nemothenoone]

Composition polynomial represents a polynomial of the form:

F(x, y_1, y_2, ... , y_k) = \sum_i (c_{2i} + c_{2i+1} * x^{n_i}) * f_i(x, y_0, y_1, ... , y_k, p_0, ... , p_m) * P_i(x)/Q_i(x).

Where:

• The term (c_{2i} + c_{2i+1} * x^{n_i}) is used for both logical 'AND' over all the constraints, and degree adjustment of each constraint.
• The sequence (p_0, ... , p_m) consists of the evaluations of the periodic public columns.
• The term f_i(y_0, y_1, ... , y_k, p_0, ... , p_m) represents a constraint.
• The term P_i(x)/Q_i(x) is a rational function such that Q_i(x)/P(i) is a polynomial with only simple roots, and its roots are exactly the locations in which the constraint f_i has to be satisfied.

Parameters deduction:

• (c_0, c_1, ...) are the random coefficients chosen by the verifier.
• The values of (n_0, n_1,...) are computed so that the degree bound of the resulting polynomial will be air->GetCompositionPolynomialDegreeBound().
• The functions (f_0, f_1,...) are induced by air.ConstraintsEval().
• The mask (for evaluations on entire cosets) is obtained from air.GetMask().

Composition polynomials are supposed to be used both to evaluate F(x, y_0, y_1, ...) on a single point, and on entire cosets using optimizations improving the (amortized) computation time for each point in the coset.