sample_laplace_distribution should produce a valid result on 0.0

Destroying the signal with probability 1 in 2^-53^ isn't a great idea. Let's pick a sensible double value, and pass it through the function instead.

I suggest 2^-54^, but it really doesn't matter exactly what value we use, as long as it produces valid results, because the probability is so low.

Introduced in 45bc5a07