Stop persisting forecasting model artifacts by default

[BelhsanHmida]

Problem

The forecasting pipeline currently saves trained model .pkl files to a local artifacts directory by default, which can fill up local directories over time.

Current behavior

• model-save-dir defaults to flexmeasures/data/models/forecasting/artifacts/models
• the training pipeline always writes the model file there
• cleanup exists via an internal delete_model flag, but it is not exposed to users

Suggestion

Change the default behavior so trained model files are deleted after prediction unless persistence was explicitly requested.

Follow-up

We could expose the existing cleanup/persistence behavior as a user-facing parameter, for example save-model or persist-model.

Questions

• Should this be exposed only as a CLI flag, or also via the API/config?

[Flix6x]

We need a plan for how to make sure the size of the artifacts folder is kept in check. We should not rely on API users to clean up after themselves. I could imagine hosts could run a CLI command in a daily cron job that cleans up older files from the folder, or maybe each forecasting job does this as a chore. There are probably other designs possible we may want to consider. Generating a list of considerations itself would be helpful as a first step, too.