Maximize acquisition via NLP

[cnpetra]

• expose derivatives in acquisition function maximization (EI and LCB evaluation and gradient) to be used with quasi-Newton algorithms
• Python interface of Ipopt (use quasi-Newton)
• Update user interface class to support analytic constrains (including Jacobian)
• Handle analytic constraints as "hard" constraint in the acquisition function
• include slsqp (current NLP solver) if supports derivative-based optimization.

[cnpetra]

closed by pr #722