Add in methods to scale the problem objective, constraints and their derivatives for IPOPT

[moorepants]

IPOPT converges better if the objective and constraints are scaled. We can probably make smart scaling based on the dynamics that are being modeled.