Add an option to specify that a Jacobian has a symmetrical structure

[amontoison]

We don't need to compute and store the CSC format of J' when we build the bipartite graph with this property.
An example is the arrowhead matrix or jacobians in ACOPF problems.

[gdalle]

Why are ACOPF Jacobians symmetric?

[amontoison]

The are not symmetric in terms of value, just for the sparsity pattern.
I think we did that for this application:
https://www.gerad.ca/papers/G-2021-56.pdf

[amontoison]

@gdalle I think we can add a field structural_symmetry::Bool in the GreedyColoringAlgorithm and specify that it's only used for coloring using a bipartite graph.
The idea is to just be able to easily pass this boolean to the constructor of bipartite_graph.
It can be relevant when we want to do a row or column coloring of a Hessian (which is cheaper than star or acyclic coloring).
For example, on symmetric banded matrices, it probably requires the same number of colors as the number of bands.
Analogously, a Jacobian can have symmetric sparsity pattern with non-symmetric values.

function bipartite_graph(J::SparseMatrixCSC; strutural_symmetry::Bool=false)
    g2 = Graph(J)  # columns to rows
    g1 = strutural_symmetry ? g1 : Graph(sparse(transpose(J)))  # rows to columns
    return BipartiteGraph(g1, g2)
end

[gdalle]

I think it might be needlessly confusing for users to have the option of doing a :nonsymmetric coloring on a matrix with "structural symmetry"

[amontoison]

We need to find a better name for the option in that case.
But symmetric sparsity pattern with unsymmetric values is quite common in linear algebra.