Parameterize hadoop wfas; change master flow setup; check master flows.

First, we add key-val parameters.  broker_name=foo sets the broker name
to foo for registration and lookup (defaults to 'hadoop').
runhadoop=False (or other False equivalents) will force the wfas to skip
Hadoop config and run.

We also separate out master<->slave flow setup and put it at the
beginning of the allpairs flow mesh setup; this was for debugging.

Finally, we also try and retry a simple parallel ssh task from the
Hadoop master to all slaves, to "verify" and "wait" for them to all be
up and reachable.  This helps us catch problems like when a VM can't
fully boot because it failed to get its metadata, or whatever.
1 job for master in 2 seconds (queued for 1 second)
Status Job ID Name Coverage
passed #773