Instead of calculating the number of rows/columns in a SparkDataFrame on the SparkR side we let SystemML handle it
Matrix block size is not passed in so the default can be configured via the SystemML settings
R4ML matrix does not cache by default (since it is no longer needed for accurate count)
This PR decreases the time required to convert a 55 GB r4ml.matrix to a SystemML matrix by 69%
Tested with the following configuration: 4 executors / 90 GB executor memory / 40 cores per executor
This PR decreases the time required to convert a 55 GB r4ml.matrix to a SystemML matrix by 69% Tested with the following configuration: 4 executors / 90 GB executor memory / 40 cores per executor