fix: construct retrained Solution with valid fields

[Tyagiquamar]

Problem

Retraining mode (--is-retrain, exposed through main(..., is_retrain=True) -> retrain_model) always crashes with:

TypeError: Solution.__init__() got an unexpected keyword argument 'iteration'

at the very end of the run - after the model has already been fully trained and packaged - so users lose the entire run.

Root cause

The Solution dataclass identifier field was renamed from iteration to solution_id during the workflow rewrite (#161), but retrain.py was never updated. The constructor call at the end of retrain_model still passes iteration=0, which is now an invalid keyword argument. Because retrain_model wraps everything in a catch-all except Exception, this surfaces as a generic RetrainingError after all expensive work has completed.

Fix

Construct the returned Solution with the current required fields (solution_id=0) in plexe/retrain.py.

Testing

• New test tests/unit/test_retrain.py runs retrain_model end-to-end against a fabricated xgboost model package with a mocked training runner (no Spark or LLM keys needed). On main it fails with TypeError ... unexpected keyword argument 'iteration'; with this fix it passes.
• python -m pytest tests/unit -q -n 2: 107 passed, 15 skipped
• ruff check and black --check: clean on changed files

[Tyagiquamar]

Hi @marcellodebernardi, gentle review ping when you have some bandwidth. #194 and #193 are both small focused bug fixes with regression tests and have been ready since Aug 25. Happy to address any feedback. Thanks!