[C++] Remove next_point argument from GradientDescentOptimization()

[suntzu86]

At some previous point, I intended to make it so that callers communicate with optimizers by doing:

• objective_state.SetCurrentPoint() to set the initial guess (pre-call)
• objective_state.GetCurrentPoint() to fetch the solution (post-call)

Somehow I missed GradientDescentOptimization in gpp_optimization.hpp. Let's remove this superfluous argument and update the code & the call-sites accordingly.

NewtonOptimization (same file) is an example of a function where this has been done.