Resolve Gaussian process predictive model

[ericward-noaa]

We've implemented GP predictive process models in the existing code. This uses the cov_exp_quad() function for the latent GP process (at knot locations) -- but we also create a covariance matrix to project to the data locations, and that doesn't use cov_exp_quad(). This may be sped up a little, and it might be worth speed testing a full rank model - to see if the predictive process should just be dropped

[ericward-noaa]

cov_exp_quad() implemented for both. Not a huge speedup, and convergence is sensitive to prior, as described here: https://betanalpha.github.io/assets/case_studies/gp_part3/part3.html