Add correlation-based localization options

[iangrooms]

This PR adds two correlation-based localization functions: one based on generalized cross-validation (preprint coming soon), and a 'prior-optimal' localization scheme from Menetrier et al. (2015) (see also Morzfeld and Hodyss (2023)). They are qualitatively similar to Anderson's Sampling Error Correction (2012) but without the need for pre-computation and a lookup table. The new options can be enabled by adding a "correlation" line to the config.yml file. E.g.

localization_def:
  horizontal:
    type: gaspari_cohn
  vertical:
    type: gaspari_cohn
  temporal:
    type: exponential
  correlation:
    type: gcv

The correlation-based schemes can be turned off with either "correlation: null" or "type: null". Generally these should be used in combination with distance-based localization, which they tend to make more robust to changes in the localization radius.

The PR also makes two other very minor changes:

• It updates the EAKF regression algorithm to avoid un-necessary divisions by N-1. Some variable names are changed from var to ss for 'sum of squares.' This saves a very small amount of cost and could be rolled back to increase code readability.
• Fixed a typo in the README

[myying]

@iangrooms Awesome! Thanks for filling this in, it's been an empty placeholder for a while. I also have the source code for Morzfeld NICE scheme, which I can add later.

[iangrooms]

@myying my pleasure! I hope it is useful.