verify accuracy of fractal S(q) for small q

[pkienzle]

The equations for the fractal S(q) model were complex enough that it was not immediately obvious how they would behave at small q.

This PR rearranges the equations to make the limiting cases much more apparent, and checks them against the original equations from Teixeira (1988) using 500 bit precision.

The numpy version of the equations was added to sasmodels.special so it will be available for pure python models.

[butlerpd]

The only change to GUI users here is the addition of the DOI for the reference to documentation which Should be accepted. The other changes are to lib files which as far as I know are not currently available through the GUI though in principle are available through scripting? Probably we just need a code review here?

[pkienzle]

Regarding the transformation, I multiplied $t_2$ by $(qξ)^{D-1}$ giving intermediate $t_2'' = (ξ/r)^D/(qξ)$, and $t_3$ by $1/(qξ)^{D-1}$ giving the new $t_3'$.

I then rearranged $t_1$ and $t_2''$ forming the new $t_2'$ using $\sin((D-1)\tan^{-1}(qξ))/(D-1)$ from $t_1$ and $1/(qξ)$ from $t_2''$, and forming the new $t_1'$ using $Γ(D+1)$ from $t_1$ and $(ξ/r)^D$ from $t_2''$.

Limit as $q \rightarrow 0$ for $t_2'$ and $t_3'$ are both 1, so only the constant $t_1'$ remains.

Limit as $D \rightarrow 0$ for $t_1'$ is 1. For $t_2'$ it is $\sin (\tan^{-1}(qξ))/(qξ) = 1/\sqrt{1+(qξ)^2}$ and for $t_3'$ it is $\sqrt{1 + (qξ)^2}$ which cancel.

Limit as $D \rightarrow 1$ for $t_1'$ is $ξ/r$, for $t_2'$ is $\tan^{-1}(qξ) / (qξ)$ and for $t_3'$ is 1.

Further check: Limit $q \rightarrow 0$ for $D \rightarrow 1$ equals $ξ/r$ matches limit $D \rightarrow 1$ for $q \rightarrow 0$.