Clarify the interface of non-default constructors of the Grid class

Currently, the Grid class supports two class methods to construct an instance of the class (i.e., factory methods).

• from_generator(): Construct an instance with generating_values generated from the passed generating_function.
• from_value_set(): Construct an instance with "generating values", i.e., one-dimensional generating points which will be expanded to the required dimension.

The method from_generator() generates the required "generating values" and calls from_value_set() under the hood. The method from_value_set() calls the default constructor with the passed "generating values" (which is expanded to generating_points) and sets the generating_function to None.

Acccording to the internal TODOs, the name "generator" is a problematic one in Python because it has already a specific meaning. I would rename it to generating function as it was done in the refactoring of the default constructor (Issue #71 (closed); MR !125 (merged)). Furthermore, I also would like to have another method call from_points() as an explicit method to create an instance of the Grid class only with generating_points (and generating_function is set to None).

In summary, four factory methods will be available:

• from_degree(m, n, p, generating_function, generating_points): Create an instance with a complete multi-index set, the given generating function (default to Chebyshev) and generating points (default to None).
• from_function(multi_index, generating_function): Create an instance with generating_function as the generating function of the Grid instance. The required generating points will be created on the fly. This method replaces from_generator().
• from_points(multi_index, generating_points): Create an instance with generating_points as the generating points of the Grid instance. The generating function of the instance is set to None. More points may be passed than what are required by the multi-index set, however, because the generating function is not available there is a limit to upgrading the grid to a higher degree.
• from_value_set(multi_index, generating_values): This methods remains in place; however, it is understood that "generating values" are interpolation points in one-dimension. So it must be given either as a one-dimensional array or a two-dimensional array of a single column/row.

What are the requirements of the arguments passed to these constructors?

• m, n, p follow the requirements of multi-index set constructor.
• generating_function: it must have the following (type) signature (Callable[[int, int], np.ndarray). The first parameter is the polynomial degree and the second is the spatial dimension; it returns an array of generating points of shape (poly_degree + 1, spatial_dimension). Each column of arrays produced by the function must contain unique values (no repeats).
• generating_points: it must be a two-dimensional array of the shape (poly_degree + 1, spatial_dimension) where poly_degree is the degree of the grid and must be larger than or equal to the maximum degree of the multi-index set across all dimensions. spatial_dimension must be equal to the spatial dimension of the multi-index set. All values in each column must be unique.
• generating_values: must either be a one-dimensional array or two-dimensional array with a single column or row. The length of the array must be poly_degree + 1 where poly_degree must be larger than or equal to the maximum degree of the multi-index set across all dimensions. The values must be unique (no repeats).

This issue is part of the refactoring of the Grid class (Issue #135).