feat: accessor methods to various distribution

[Qazalbash]

Hi, this PR fixes #186. I have implemented the accessor methods for the following distributions,

• Dirac
• Discrete Uniform
• Log Normal
• Triangular
• Uniform
• Multivariate Normal

[YeungOnion]

Thanks for working on this. I think it all looks good, but I'm not sure if it's best to leak the implementation of multivariate normal with precision and cov_chol

If you can make these return cloned/owned values, it will be less binding in our API - there is no implication that the struct owns such data. While I have no expectations for this to change, but we're still fairly new.

As far as cost goes, the intended matrix sizes for nalgebra are small (dim<5) will be cheap for potential alloc and copies, so an accessor with cloning is cheaper than computing cholesky decomp or inversion.

A possible intermediate for those who need the precomputed values as is: a take method that returns a struct where all fields are pub containing the same fields as the "taken" MultivariateNormal but none of statistics impls.

---

EDIT

Quoting myself in #186

non-Copy fields should not clone for immutable accessor, clones should be opt-in by caller

And I just realized that I'm giving you mixed messages on copy being opt-in, aren't I?

🤔 Hmm. I think I need someone else's input here. In what cases might you want to view the precompute outside of the features we have already implemented?

[Qazalbash]

Hi, I will keep an eye on this PR. Let me know the conclusion.