48 benchmarking fix zero and full shot

[ben-jy]

Benchmark module has been updated to handle zero shot and full shot learning evaluation. In this PR, these were done:

• restructure the benchmark module
• allow to load a checkpoint for pretrained model (especially foundation models)
• allow to retrieve some example predictions once the benchmark run is finished, useful for plotting purposes
• update and improve result dataframes
• add a wraper around simple and univariate models from darts

close #48