Making transit_poly/s2 code more efficient

[ericagol]

• [x] Reuse elliptic integrals: E(k^2) and (E(k^2)-(1-k^2)*K(k^2))/k^2.
• [x] Compute k^2, k_c, and \kappa only once.
• [x] Pre-compute coefficients of Iv_series, Jv_series, and dJv/dk series. Question: do we want to add these series coefficients to the TransitStruct structure?
• [x] Get rid of redundant division and multiplication: for instance, for s2 we compute 2\pi/3 times flux, and in computing the normalized light curve, we divide by the same factor.

[ericagol]

@rodluger I implemented the simultaneous computation of the elliptic integrals and I reused the elliptic integrals from s2 to compute I_v and J_v and their derivatives. This sped the code up by about 30% from 0.5 seconds for 10^6 transit evaluations to 0.35 seconds (for quadratic limb-darkening). So, not a huge savings, but significant, and now it's about 30% faster than Pal's code compiled with -O3. I'm doing a lot of other inefficient things in the computation, though, so I suspect that there is still more room for improvement.