pl/math: Add scalar expm1

New routine uses the same algorithm as the single-precision routine,
and is accurate to 2.5 ULP.
9 files changed