Discrete Gaussian distribution in rand_distr?

[ooovi]

I implemented a sampler for the Discrete Gaussian distribution for another project. The distribution is a discretization of the normal distribution that converges in a desirable way and allows to circumvent certain serious issues that can arise when attempting to provide differential privacy.

The implementation has been reviewed and tested against the python reference implementation, and I wrote some empirical tests as well. It is currently part of the libprio-rs crate and hence not easily reusable.

Is this the kind of distribution you think would make sense to include in rand_distr, or should i go for a seperate crate? I'm willing to make a pull request, if you're interested.

[dhardy]

I don't know how widely applicable this distribution is, and don't have the time to read into it.

At a glance, your code looks well structured, documented and tested, so I'm tempted to say yes please, at some risk of rand_distr continuing to grow as an uncoordinated and poorly maintained aggregation of stochastic samplers.

@vks do you have any thoughts?