Generating CPU family headers for multiple isas

[Ben Elliston]

   Hm, that reminds me.  Does it make much sense to have isa-specific
   hardware?  I can see some cgen-time error checking (to keep isa
   semantics limited to isa-visible hardware/operands), but at run time,
   generating a single unified struct of all isas' hardware seems
   appropriate.

Doing this also drastically simplifies the process of writing out
getter/setter cover functions to access said hardware.

Ben