Preconditioning in auto-generated solver

[alphaville]

Main Changes

Essentially, this PR finalises most of the tasks involved in #292.

In optimizer.rs:

• initialise optimiser interface
• compute scaling parameters and initial penalty
• scale auto-generated functions
• undo the scaling to return the actual optimal cost

In optimizer_builder.py:

• Functions psi, grad_psi, F1 and F2 become functions of θ=(p, w) instead of just p

In interface.c:

• Introduced a function to export the scaling parameter w_cost

In alm_optimizer_status.rs

• Included functions that allow to update the cost and the infeasibilities

Other

• Fix issue with open sockets not getting closed
• Fix issue in Halfspace (missing float)
• Testing

Associated Issues

• Addresses #106
• Addresses #107

TODOs

• [x] Documentation
• [x] All tests must pass
• [x] Update OpEn CHANGELOG (Rust)
• [x] ~Update webpage documentation~ (will do later)

[alphaville]

@Arun-R-M

A test case can be added that solves a known problem and calls the solver with initial_guess and p values and check against a hardcoded set of correct solution (possibly obtained from a previous version of opengen).

Good point, but we check whether the optimality conditions are satisfied and whether the solutions are feasible. I'll strengthen the unit test a little more.

[alphaville]

@Arun-R-M further testing revealed a bug: preconditioning scales the cost function, but then when we return the optimal cost to the user we need to scale it back (to undo the scaling). I'm pushing a change now that will make a test fail, but I'll fix it.