Skill module: Computing metrics from the hcst and obs anomalies

@nmilders @nperez

As mentioned in https://earth.bsc.es/gitlab/es/auto-s2s/-/issues/27#note_181440, it has come to light that some of the skill metrics should be computed using the hindcast and observation anomalies.

For this purpose, @nperez proposed adding a TRUE/FALSE parameter to the recipe so users can specify if they want to compute those metrics from the anomalies or not. I'm opening this issue to track the development.

As far as I am aware, we have the following options for anomaly computation:

1. s2dv::Clim() and s2dv::Ano().
2. CSTools::CST_Anomaly(). This function allows computing the anomalies in cross-validation, but it might need some minor improvements (e.g. the name of the ensemble dimension seems hardcoded to be 'member')
3. Am I missing any functions?

We should compile a list of which metrics should always be computed before anomaly computation. Here is a preliminary list:

1. Always before anomalies: Mean Bias, Mean Bias Skill Score.
2. After the anomalies: Ensemble Mean Correlation, Ensemble Correlation, RPS, FRPS, CRPS, RPSS, FRPSS, CRPSS, Brier Skill Scores, Spread-to-Error Ratio

As you have more information, please add it to the comments. Also feel free to tag any people who might have something to say in the discussion.

Cheers,

Victòria