Merging rt452_reduce_mmd branch into trunk.

[Vasily Chekalkin]

Mark Glines wrote:
> Vasily Chekalkin wrote:
>> I've changed Integer.i_ops to fallback to MMD dispatch in default case
>> (and added few MULTI methods to support it).
> 
> 
> Great.  Does that mean we can still perform awesomely in the case of
> core PMCs?  Do you think we're going to run into similar issues with any

Yes, it still perform fast. And after reducing GC pressure in last 2 commits 
it ~10% faster than before on benchmark_test.

> other PMCs changed by the branch?

Yes. For example BigInt. I'll rework it fallback to MMD in default case.

I think that current VTABLE/MUTLI/METHOD interactions is inconsistent and have 
to be re-worked...  I agree with chromatic++ comment#1 in ticket - "In the 
long term, making VTABLE calls multidispatch themselves will actually reduce 
the amount of code in the system.".

-- 
Bacek