Return value with constraints

[nikohansen]

While fmin_con2 handles the return value for x nicely, x, es = fmin2 does not. We need to access es.result for a useful solution. This should be fixed for both cases, when constraints are passed and when the fitness function is a ConstrainedFitnessAL.

Some related points

• We probably want to assign the best not based on the augmented Lagrangrian value, as this depends on the dynamically changing AL coefficients and is quite meaningless without knowing the coefficients.
• We could "extend" best_feasible to contain the solution with the smallest violation when no feasible was found yet.
• When feasible solutions were only found early on, it seems not advisable to return the best feasible solution. This also suggests a default behavior for find_feasible_final to be depending on when the last feasible solution was seen.