fixed bug (mising h in k_pivot) and better handling of zeta

k_pivot was not properly converted from 1/Mpc to h/Mpc, which let to erroneous ICs for bClass=1. The exact error introduced was a scaling in the power spectrum by a factor h^(n_s - 1) ~ 1.01. This is now fixed.

The analytic zeta function is now multiplied on the transfer functions after interpolation, not prior to. This leads to slightly better interpolations and hence more precise ICs. The cost is that zeta is now no longer computed just once (in csmClassRead), but has been separated out as its own function, csmZeta, which will be called by the csmDelta_m and csmTheta_m functions at particle IC creation and by the csmDelta_lin function at realization of the linear field.