* Extract partitioner from hist. * Implement categorical data support by passing the gradient index directly into the partitioner. * Organize/update document. * Remove code for negative hessian.
