Added code so that Spark jobs can be launched from outside the Spark

directory by setting SPARK_HOME and locating the executor relative to
that. Entries on SPARK_CLASSPATH and SPARK_LIBRARY_PATH are also passed
along to worker nodes.