realML primitives

• realML.matrix: fast least absolute deviations regression, fast robust (entrywise-l1 norm) low-rank matrix decomposition, sufficient dimensionality reduction (non-linear low-rank factorization)
• realML.kernel: preconditioned Gaussian and Polynomial regression, tensor machines for adaptive polynomial features