WL#2385: Multi-source testing

SUMMARY
  - Test of multiple channels                                     
  - Rewrite of test scripts for multiple hosts
  - Remove mi slave options from master-test-run.sh
  - Add CHANGE MASTER to master-slave.inc
  - Fix so that new filenames are used
  - Make it possible to define test connections    
    in the test script itself

TEST CASE: Hub and Spokes
-------------------------
JD writes: 
> I think we'll see people wanting to a run single slave with hundreds of
> masters in a sort of "hub and spokes" topology.
>
> Some examples might be a point-of-sale system, where a server at a
> central office replicates from hundreds of stores.  Or a monitoring
> system with one central server replicating from multiple data
> centers/oil rigs/etc with one server each.
>
> If you built this thinking of two or three masters and then people run
> it with hundreds we might see problems like too many file descriptors,
> too many threads, etc... so questions come up like: do we get good error
> messages in these situations?  Is the whole thing actually manageable?