This is a dubious usage of random number generators, if the fuzzer is supposed to test handling for special values. E.g. f64 encodes +0 when all 64 bits are zero (and -0 with first bit only 1); the chance of this is 1 in 2^64 or approx. double that, 10^-19, chance of generating any 0 (+ or -); i.e. you could run the fuzzer a billion (10^12) times and still only have a 1-in-ten-million chance of generating 0. And there's other special values you might want to test too; +/- infinity (also 63 fixed bits) and NaN (this one is much more likely, since it only fixes the exponent bits and requires non-zero fraction part).
Edit: I'm using the old-world billion of 10^12 here not 10^9. Not that it matters much, practically speaking.