Add fns Uniform::min, max using trait UniformSamplerRange

[dhardy]

• Added a CHANGELOG.md entry

Summary

Add fns Uniform::min, max using trait UniformSamplerRange

Motivation

I wanted Uniform::max and realised it's simple enough to add.

Note: some rand_distr distribution objects already support querying parameters (example).

Details

trait UniformSamplerRange: UniformSampler is an extension-trait, adding optional support for min and max. This doesn't technically need the UniformSampler, though we do used its assoc. type X.

We could use a more generic name (like HasRangeBounds or HasMin and HasMax traits), but UniformSamplerRange fits in nicely (given the bound).

Adds min, max fns on Uniform for types supporting UniformSamplerRange.

Implements UniformSamplerRange for UniformInt samplers only.

I looked briefly at the other samplers and we certainly could implement these too, but code is less trivial. In particular, UniformChar would involve duplicating unsafe code. I suggest we leave these off for now and see if anyone else requests them.

[vks]

Why are these commented out?

[dhardy]

Because the macro input uses both integers and Simd values. We'd need a generic Simd::splat supporting both... possible, but not in std I think so we'd need extra utility code (I started on this before deciding it's probably better not to increase the code size without good reason).

[newpavlov]

Do we need the trait? I think these could be just inherent methods. I am also not sure about max/min returning splatted SIMD values. Some may expect to get scalar values instead (i.e. u32 instead of u32x4). It also would resolve the assert problem.

[dhardy]

Do we need the trait?

If we only implement these on UniformInt probably not. If we want the methods on Uniform too, I think we do (unless we fold them into UniformSampler, but that would be a breaking change and require extra impls).

I am also not sure about max/min returning splatted SIMD values.

UniformInt stores SIMD fields internally; we don't require range inputs to be splatted scalars.

[dhardy]

Given the low interest I guess I'll just close this (it's not an essential addition).