This is a work in progress.  Updating the floating point comparisons
require changes in AST, IR, and all backends.  Seems useful to break
the overall changes in to a couple of commits.  This change is focused
on getting the new AST in place, though includes some llvm/gas64/gcc
backend updates and leaving the gas/gas+SSE for a later update.

- handle IR_EMITOPT_REL_DOINVERSE option in gcc/llvm/gas64 backends
- gas64 x86_64 backend : re-use astGetInverseLogOp()
- gcc backend all targets: use !(float op float) to implement reverse logic
- llvm backend: Add floating point comparison instructions,
  Note: currently no testing completed for NaN's in LLVM
- 'fpmode fast' will disable NaN checking - and will affect
  all backends.  Comparisons involving NaN's will be inaccurate
  however, all other comparisons of numbers (including infinities)
  should still be correct.
- idea is to let the backend optimize float comparisons where possible
- otherwise use safer code generation to get consistent results across
  branching, IIF, expressions, etc.