tabulated sersics

Based on a discussion with @rmjarvis and @esheldon today, it would be useful to have ways to circumvent the Sersic profile limitation that there is a significant setup time for every single new value of n. There were two things we discussed, one easier and one harder:

(1) Easier: If one has a p(n) they want to draw from, we could offer some discretization option to DistDeviate so that instead of varying n continuously, it draws from p(n) at some specific number of discrete values. Presumably if you tell it you want e.g. 100 different values from p(n) then it should choose those values to be evenly-spaced in the CDF (e.g., at p=0.01, 0.02, 0.03, ...) rather than evenly-spaced in n.

(2) Harder: we could tabulate the Hankel transforms that are the main setup cost, for some number of n values, and just interpolate between them. There is a tricky research problem here, which is to figure out what resolution in n is needed to get systematics below the desired level. Actually doing the tabulation and having code that can read them in shouldn't be hard.

Start by reading the DistDeviate behavior and the Sersic profile's Hankel-transform setup. Compare the proposed discrete p(n) sampling with tabulated transforms and interpolation, including the required n resolution and systematic-error target. Done requires a defined approach and implementation plan that addresses the setup-time cost.