Add newton optimization method

[ShkalikovOleh]

1. Unified point history structure
2. Add to step and direction history for descent methods
3. Add step optimizer abstraction(and rewrite all descent methods via this idea)
4. Add base newton and quasi newton methods class
5. Add simple newton method. Direction is dot product of inverse hessian and gradient

[ShkalikovOleh]

Now all optimization algorithms return OptimizeResult(dict-like) object