Disable -mfpu=neon for aarch64

[hesmar]

This PR removes the -mfpu command line argument on aarch64, because it is not available in gcc on this architecture.
See: https://gcc.gnu.org/onlinedocs/gcc/AArch64-Options.html

[stweil]

I'd prefer to add this code at move the whole if block to line 262 because I think that looks simpler and clearer:

  check_cxx_compiler_flag("-mfpu=neon" HAVE_NEON)
  if(HAVE_NEON)
    set(NEON_COMPILE_FLAGS "-mfpu=neon")
    [...]
  endif

[hesmar]

At the moment, I only moved the NEON_COMPILE_FLAGS up because I am not sure if it is safe to set the ENABLE_LTO only for arm. If you say that this is safe, I can update the PR again.

[stweil]

The comment says that link time optimization had a problem on arm/RBPi, therefore I think it was only for 32 bit ARM. Disabling LTO for all aarch64 builds does not look reasonable for me. But that can be handled in a separate pull request.

[stweil]

Looks good, thank you.