Enable testing unconstrained solvers on constrained problems

[zaikunzhang]

The idea is to test as many problems as possible.

Consider a constrained problem

$$\text{minimize} ~ ~ f(x) ~~\text{s.t.}~ ~ c(x) \le 0.$$

There are two ways to test an unconstrained solver on it.

1. Ignore the constraints completely. The problem may not be well-posed anymore. However, it does not hurt to benchmark unconstrained solvers on it and see which solver can decrease $f$ more at less cost.

2. Penalize the constraints. The possible penal functions are Courant (quadratic) and L_p. The penalty parameter should also be configurable.

Note that this may produce repeated problems, which I do not think is a big issue.

In the same way, we can also test bound-constrained solvers on linearly constrained problems, etc.

Thanks.