Add support for monomial evaluation

[moralapablo]

Current implementation of eval_poly() provides the final value of the given polynomial. However, to compare with typical XAI methods that provide feature importance in a local manner, such as SHAP values for a given observation, it would be nice to be able to compute the value of each monomial in the polynomial for a given observation.

As an example, having the polynomial $3+ 2x_1 - x_2 +4x_1x_2$ evaluated on $x=(1,1)$, would be $3 + 2 - 1 + 4$, where each of those values is the contribution of each monomial evaluated on the given point $x$.

Output dimension discussion

Currently, eval_poly() receives poly$values as a matrix, where each column represents a polynomial (all with the same terms) and each row is a given monomial in all those polynomials. It also receives a newdata item that can be a matrix or a dataframe where the columns are the variables or features and the rows the observations or instances.

Then, the output of eval_poly is built as a matrix with polynomials as columns (matching the columns in poly$values notation) and observations as rows (matching the rows in newdata notation).

However, a function that can compute monomial values as suggested needs to have a 3D output, as it would need to keep the different polynomials and observations dimensions as before, but also the monomials dimension, given by the rows in poly$values. An array with 3 dimensions seems the best option, where the dimensions could be c(n_observations, n_monomials, n_polynomials), so for the polynomial k, element [ , ,k] will be a matrix with observations as rows and monomials as columns.