cluster metadata should be supervised separately

[jrwest]

for several reasons:

a) isolation -- the usual benefits of a more granular supervision tree
b) testing -- have a "test" mode that starts all the parts in a convenient way for unit/eqc testing.

[jrwest]

marked as milestone 2.0.1, small improvement to existing feature, not necessary for 2.0-RC