Update Gaussian random shear field generation

[rmandelb]

@gbernstein pointed out that this paper: http://arxiv.org/abs/1406.6072 has an algorithm for making Gaussian random fields that isn't subject to some of the issues that show up with our current periodic box-dependent algorithm, such as failure to reproduce shear correlations on scales comparable to ~1/2 the size of the simulated area. We should consider implementing this in GalSim.

Thanks for pointing this out, Gary!