New procedures for floating-point fused-multiply/add:

- (flo:*+ u v a) computes u*v + a correctly rounded with no
  intermediate overflow or underflow
- (flo:fma u v) is the same, with a more familiar name
- (flo:fast-fma?) returns true if fma is hardware-supported and false
  if it is emulated in software with Dekker's double-precision
  algorithm
