Issue with Von Mises-Fisher sampler returning NaN points

[Kevin-Mattheus-Moerman]

TLDR: I believe #1930 by @devmotion might fix my issue so I was wondering when you will mint a new release so I can include this package as a dependency in one of my own.

Thanks for making this awesome package. I've just discovered the package today. I have been trying to create and use samplers for VonMisesFisher distributions. I could not find anything in the documentation but based some things on the source and tests here. I tried the following for instance:

μ = [1.0, 0.0, 0.0]
κ = 50.0
spl = VonMisesFisher(μ, κ)
a = rand(spl)

However, this produces a as:

3-element Vector{Float64}:
 NaN
 NaN
 NaN

Also it seems:

spl = Distributions.VonMisesFisherSampler(μ, κ)

Features spl.v as:

3-element Vector{Float64}:
 NaN
 NaN
 NaN

This seems to relate to μ = [1.0, 0.0, 0.0] as other choices of μ did not seem to produce this issue.

When I run the latest release I get the above issue, but I cannot reproduce the issue when I clone the latest development version.

---

Side notes not relevant to issue:
FYI I'm the developer for Comodo and will be using this to create Von Mises-Fisher mixture models:

Which define wave vector directions to help create something called spinodoid structures: