Fix Cosmology.from_sigma8() for a_stop!=1

[pointeee]

In #30 Adrian mentioned that current growth function is tabulated between a_start and a_stop, and the growth function gives wrong values for input scale factor out of the region because of the extrapolation behavior of jax.numpy.interp. This can be fixed by always tabulating the growth from a=0 to a=1 in the configration.

[adrianbayer]

Previously the code only calculated the growth at the exact a values of the nbody sim, but now it computes it at an arbitrary linspace. Would something like jnp.concatenate((self.a_lpt, self.a_nbody[1:], [1])) be better? I.e. just adding [1] as an extra concatenation so that it's the same as the old code, but also with a=1.