See #198 for details.
Basically, our BigDecimal defines "new" instead of an allocator and initialize method. Initialize is never called, and until my fix for #198, we did not observe subclasses at all. We need to correct this to allow subclassing BigDecimal.
basic BigDecimal sub-class test ... all working the same as MRI 1.8/1.9
... no need to address #365 as the allocate+initialize sequence seems MRI-like
it is actually how 1.8.7 / 1.9.3 behaves with BigDecimal sub-classes (does not invoke initialize)
... but the bug has been open for long enough to be ripe for 9K since 2.2.2 does call initialize :)