Add compiler flags to control the LLVM compiler optimisations

As the title says.

This really comes in two stages

• COMPLETED -O
  • Handle one flag, -O which is like that of, e,g., GHC, GCC, where you control a curated set of passes
  • LLVM has such curated pass sets, so this is the simplest stage to support
• -f
  • -f flags, e.g., -ffast-math, are much more general, being switches for each particular optimisation
  • this one seems much harder to do

Another point of note is to see if this -O0 flag would be equivalent to the --no-llvm-opt flag (#266)