Improve comments slightly (#135)

* Strip trailing whitespace.

* Change __builtin_ctz comments.

In f2s.c and d2s_intrinsics.h, explain that the 32-bit and 64-bit
flavors don't appear to be faster than the existing code.

In generic_128.h, simply drop the comment, as there is no 128-bit
builtin as far as I know.

* Fix comment typos.