A bunch of Hadoop config changes.

Let the master node just be the resourcemanager (thus in the current
test setup, the resourcemanager node does nothing).  Don't make the
master a slave (datanode) node -- this seems to force HDFS to spread
input files around to all datanodes instead of keeping it locally on the
master.  Also don't tell slaves about the slaves list file.
1 job for master in 2 seconds (queued for 2 seconds)
Status Job ID Name Coverage
passed #771