Fix polar angle sampling and Mott factor in Wentzel distribution

[amandalund]

This fixes what I think are a couple small errors in the Wentzel distribution:

• In the sampling of the scattering polar angle, looking through @hhollenb's (excellent!) Coulomb scattering writeup, I think the sampling function for mu (Eq. 2.10) should be
  $$\mu = \mu_1 + \frac{(A + \mu_1) \xi (\mu_2 - \mu_1)}{A + \mu_2 - \xi (\mu_2 - \mu_1)}$$.
  With $\mu = \frac{1}{2}(1 - \cos \theta)$ and $\mu_1 = \mu(0) = 0$, this gives
  $$\cos \theta = 1 - \frac{2 A \xi \mu_2}{A + (1 - \xi) \mu_2}$$.
  The sampled angle should now always be in [-1, 1), so we can remove the clamp.
• In both Geant4 and the writeup (section 8) the Mott factor is given as $1 + 2\times 10^{-4} Z^2$. Updating this did get rid of an assertion failure in the rejection sampling where the p_true in the Bernoulli distribution was greater than one.

It would be good if @hhollenb could confirm that these changes are correct and I'm not missing something.

[sethrj]

Nice find! Good to justify our general avoidance of clamp and its ilk...

[sethrj]

Darn it, I guess we'll never replace multiple scattering with single coulomb at this rate...

I'm happy either to wait for @hhollenb 's double check before merging this or to let you go ahead and merge yourself.

[hhollenb]

This looks great, and may fix some of the small discrepancies in the scattering angle.

I used clamp out of a concern for other parts of the code requiring $|\cos\theta| \leq 1$ and there being small rounding errors. More than happy to replace it with an ensure post-condition.

I'll approve it and merge it into the tests I have right now. Finding that source for @amandalund coulomb limit made me want to revisit the weird false scattering code and make sure its assumptions are correct.

[hhollenb]

I'll revisit the Coulomb scattering limit with Soon once he returns and I review the paper.