Use coordinate descent for continuous variables

[nicoladicicco]

This draft PR implements coordinate search for continuous variables in place of the current random search.
The objective function is assumed to be $C^1$ w.r.t. the continuous variables. Gradients are (momentarily) estimated via a simple finite difference method.
The algorithm uses the Armijo line search criterion to determine the step size.

The following features should be implemented before marking for review.

• Implement a basic line search algorithm for coordinate descent on continuous variables
• Integrate the coordinate descent algorithm with the management of the Tabu list
• Add a test for a continuous problem (e.g., chemical equilibrium from ConstraintModels.jl) and check for convergence. The test should fail on the current version of the package

[Azzaare]

Regarding ConstraintModels, I still need to update the package!

Thank you for the nice work on this PR @nicoladicicco