Optimization and reliability improvements

[rjagerman]

This pull requests adds numerous features and improvements:

• Many optimization improvements
• Refactored and simplified partitioning scheme with new unit tests
• Push and pull requests now go through a finite-state-machine with additional communication logic
  • Push requests guarantee exactly-once delivery
  • Pull requests guarantee at-least-once delivery
• Add a RetryBigMatrix and RetryBigVector to have an extra layer of retry functionality during network failures
• Added pipelined iterators for easy-to-use and high-performance iteration over entries of the matrix:
  • ColumnIterator: Iterates a BigMatrix column per column
  • RowBlockIterator: Iterates blocks of rows of a BigMatrix
  • RowIterator: Iterates a BigMatrix row by row (uses underlying RowBlockIterator as a buffer)

[rjagerman]

Tests on travis are failing due to akka timeouts on the GranularBigMatrix test, which is quite large. I will reduce the test size soon (while also reducing maximum message size in the akka config so we still test for the correct behavior) and update the pull request.