RFR: Make sure we have at least one memory pool per memory manager (JMX)

[Roman Kennke]

Hi,

this adds a memory pool to the major memory manager in Shenandoah's JMX
interface. I suspect that this was actually meant to be that way, but we
made a mistake and assigned the 2nd pool to the minor mgr.

It *is* possible that we assigned two pools to the minor mgr on purpose,
because the services interface is so wretched, and certain clients
assume a fixed amount of memory managers and maybe pools. It doesn't
really fit what Shenandoah does though.

This patch fixes a failure in the TCK. I wrote a testcase that verifies
kindof the same thing as the failing TCK test (i.e. that each memory mgr
has at least one memory pool with non-empty names, etc). Whenever some
other implicit requirement of clients shows up (e.g. need exactly N
pools or managers) we must add testcases to verify this.

Tests: hotspot_gc_shenandoah, manually running jstat

http://cr.openjdk.java.net/~rkennke/jmxmempools/webrev.01/

Ok?