Refactor objective, gradient and evaluate_fun_with_grad methods

[jhelgert]

This small PR is motivated by the discussion here. It refactors the evaluate_fun_with_grad, objective and gradient methods in the scipy interface and handles the jac=True option for the objective function similar to the constraints by using the MemoizeJac decorator.

Should we leave the evaluate_fun_with_grad method as it is, or do you think deprecating it is the better way?

[moorepants]

Can you provide more information? I'm not sure what this is doing or what you want to deprecate.

[jhelgert]

Sure! While discussing my last PR, @brocksam proposed to refactor the relevant parts of the evaluate_fun_with_grad() method into the objective() and gradient() methods. The constraints already use the MemoizeJac Decorator:

cyipopt/cyipopt/scipy_interface.py

Lines 107 to 118 in dc17c81

	for con in constraints:
	con_fun = con['fun']
	con_jac = con.get('jac', None)
	con_args = con.get('args', [])
	con_hessian = con.get('hess', None)
	con_kwargs = con.get('kwargs', {})
	if con_jac is None:
	con_jac = lambda x0, *args, **kwargs: approx_fprime(
	x0, con_fun, eps, *args, **kwargs)
	elif con_jac is True:
	con_fun = MemoizeJac(con_fun)
	con_jac = con_fun.derivative

So there's no need to check whether the constraint function returns both the constraint function value and the jacobian inside the constraint() and jacobian() methods.

This PR does the same for the objective. That is why we no longer need to call evaluate_fun_with_grad() inside objective() and gradient() since the case that the objective returns a tuple containing the objective value and the gradient is already handled by the MemoizeJac decorator. I hope this makes things a bit clearer.

[moorepants]

I think we can just leave the evaluate_fun_with_grad() method and not worry about trying to deprecate.

[moorepants]

It isn't clear why this is removed.

[jhelgert]

This should have been removed in this commit in the last PR. Sorry for the mess!