replacing `expm1` implementation

`expm1(x::Double64)` loses all digits of accuracy below `abs(x) < 1e-163` and returns `0.0` instead of `x`.

This regresses even beyond Float64, which correctly returns `x`.

```
using DoubleFloats
x = 1e-200
@assert x == expm1(x)
xx = Double64(x)
@assert xx == expm1(xx)
```