cmmToCmm should be the first class CmmToCmm pass

[Donald Bruce Stewart]

Hey,

The cmmToCmm pass that lives in nativeGen/AsmCodeGen seems to do some
useful things. These optimisations seem to be just as useful in the C
backend, I think. The constant folding certainly makes the C output look
smarter.

Should this be moved out of the nativeGen, and into, say, cmm/CmmToCmm,
as a generic cmm optimiser?

-- Don