Improve dithering CPU usage

[roderickvd]

This little PR shaves off about 33% CPU usage when dithering, while the effects on the binary size are next to none. This is true both on my Raspberry Pi 3B+ (from about 15 to 10% CPU) and my Mac (from 3 to 1%).

The trick is taking cue from SmallRng instead of ThreadRng, then sprinkling some 64-bit optimisations on top by hand-picking the fastest generator that still has high enough linear complexity for 64-bit floats.

SmallRng may be a good choice when a PRNG with small state, cheap initialization, good statistical quality and good performance are required. [...] The PRNG algorithm in SmallRng is chosen to be efficient on the current platform, without consideration for cryptography or security.

While at it, bump the rand_distr crate, even though it doesn't contribute to the speed-up.

Unless there are any objections, I intend to merge this before releasing 0.3.1.