RFR: Initial committed regions count can be more than max

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/committed-max/webrev.01/

This hot-fixes the apparent off-by-one error in initial region count calculation. We normally deal
with rounding errors by dealing with capacities in "region" units. That works well, but we also need
to make sure the initial number of regions does not get more than max number of regions.

Separately, I shall look into the underlying issues why we have less "max" regions than "initial".
The hotfix makes the downstream code correct meanwhile, and it makes sure failures are not cryptic,
but actually asserted before pushing bad values down.

Testing: tier3_gc_shenandoah, failing ad-hoc tests

Thanks,
-Aleksey