A library to support distributed matrix computation for machine learning and data analysis

New branch of MatFast project. The code base moves from Spark-1.5.0 to Spark-2.1. We have changed the project name to "MatFast".

Now, We are integrating the matrix operations and optimizations inside Spark SQL. We have extended the DataSet API such that all the implemented matrix operators are available to Spark SQL. We are also working on extending the Spark SQL Catalyst such that the optimizer can automatically recognize the special structures of matrix queries.

This work was accepted by ICDE'17. More technical details and design decisions can be found in the paper "In-memory distributed matrix computation processing and optimization". We'll continue updating the code base to embrace more useful features.