`ValueError: could not convert string to float` when training some models

[marco-c]

We're seeing:

Traceback (most recent call last):
  File "/usr/local/bin/bugbug-train", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/usr/local/lib/python3.12/site-packages/scripts/trainer.py", line 143, in main
    retriever.go(args)
  File "/usr/local/lib/python3.12/site-packages/scripts/trainer.py", line 41, in go
    metrics = model_obj.train(limit=args.limit)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/bugbug/model.py", line 424, in train
    explainer = shap.TreeExplainer(self.clf.named_steps["estimator"])
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/shap/explainers/_tree.py", line 278, in __init__
    self.model = TreeEnsemble(model, self.data, self.data_missing, model_output)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/shap/explainers/_tree.py", line 1261, in __init__
    self._set_xgboost_model_attributes(
  File "/usr/local/lib/python3.12/site-packages/shap/explainers/_tree.py", line 1506, in _set_xgboost_model_attributes
    loader = XGBTreeModelLoader(self.original_model)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/shap/explainers/_tree.py", line 2104, in __init__
    self.base_score = float(learner_model_param["base_score"])
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: could not convert string to float: '[5E-1]'

when training the stepstoreproduce and the test failure models.

Maybe a regression from #5332 or #5376?

[marco-c]

Reverting them didn't help (see #5425 and #5426), so it must be something else.