Fix inverse_cdf for Gamma

[YeungOnion]

Don't think there's a closed form expression for inverting the CDF, so this improves on the bisection approach with a few iterations of Newton-Raphson since we have pdf

also adds tests

[FreezyLemon]

Is there a special reason for defining the number of iterations like this? It's probably more natural to just

for _ in 0..8 {

since the complexity is pretty low

[YeungOnion]

I thought,

8 and 4 sound like nice numbers, what does that give me. 8 bisection is 2 digits in $p$ and NR is quadratic convergence, so 4 gives 16 digits, lucky me. Hmm, it's helpful to name things, so const it is.

Then I thought,

I don't want to name two close but separate things, tuple it is.

I.e. I would be fine changing it to literal.

However, since my attention is on it, I think I might instead do the initial bracketing a little differently and try some math to back up the number of iterations I select.

[FreezyLemon]

I see. It's not a problem either way, I just wanted to know why you used that const tuple there.

[YeungOnion]

Since it works and I've not already reviewed how to make it "better" I don't think I will for now.

If we get a handle on provided or emitting precisions then I think property testing shrinking could use some of that precision information to help find larger parameter space with correct behavior.