Surrogate optimizer based on gradients

[Balaji Pokuri]

Since both mean and variance can be calculated analytically, there should be provision to find optima of surrogate based on its gradient, at-least the first derivative. This can enable finding multiple surrogate optima with the same acquisition function / exploration-exploitation mode.

[Balaji Pokuri]

Can also help to avoid repetition of next_query_point across multiple acquisition functions.

[Balaji Pokuri]

gradients implemented in 5d97dd1 . Integration with optimizer should be done. Can be combined with Issue #31

[Balaji Pokuri]

derivatives of acquisition function should be implemented for it to be used effectively