Modularize Optim.jl

Wrap everything in module block
Export optimize function.

Priority Changes

Set up Nelder-Mead to automatically create points around a starting vector.
Add even more default variations.
Get benchmarks to run on all of the problems provided.
Make optimize() the best function to use.

Step-Size

Add more complex step-size method from Nocedal and Wright.

General Changes

Switch over to using base/distributions.jl for all RNG's

Method Changes

Revise L-BFGS
Add L-BFGS-B
Add all methods from Boyd and Vanderberghe's Convex Optimization book
Add all methods from Nocedal and Wright's book
Incorporate conjugate gradient code or write from scratch.

Calculus.jl Changes

Write function to estimate the gradient of a function numerically
Write function to estimate the Hessian of a function numerically

Docs Changes

Improve documentation

Testing Changes

Improve tests using both 1D and 2D problems
Need to get ridge regression example to produce same solution as glmnet() in R
All methods should be tested and timed on:
- Rosenbrock
- Powell's
- Simple parabola in 5D
Track run time (after initial compiling run), number of iterations