Convert all NLopt calls to use Optimization.jl wrapper

[jlperla]

#56 first.

At this point, it would require adding in new features to the GalacticOptim to support equality constraints.

There should be two places:

• amss.md
• opt_tax_recur.md

Both use the min_objective and we can keep it that way for now, but ideally we would add support for max_objective if it is more natural in this example.

[jlperla]

While equality constraints are not supported in the Optimization.jl wrapper, we map be able to port these with inequalities. The reason is that the algorithms used in AMSS don't directly support equality, and internally convert to inequalities---so it should be without loss. Need to see if other equality constrained optimizers are used.