Enable BiomeE likelihood for calibration with a) multiple targets and b) multiple sites

[fabern]

BiomeE likelihood function was very basic, mainly a hardcoded proof-of-concept but not flexible enough to apply to multiple sites or multiple target variables.
This PR updates this.

This PR further serves as a opportunity to streamline the likelihoods/RMSE calculations and properly applying the DRY principle (reducing the repeated code).

• add tests for current behavior
• homogenize likelihoods for p-model and BiomeE-model
• split likelihood function in
  • a first part that runs the model and generates an obs_pred_data.frame
    • TODO: streamline pred_obs result from likelihoodHelper_combine_model_obs() into a single, long data.frame()
  • and a second part that computes RMSE/likelihood/GoF between obs and pred
• same as above for RMSE functions, too
• add argument verbose=FALSE from PR: Minor changes #217
• update vignettes, tests etc. with the split-function-approach

This is not only relevant for BiomeE but also changing the calibration of p-model to use ET instead/in addition to GPP. Hence this is also related to: geco-bern/sofunCalVal#5 and geco-bern/sofunCalVal#17