Bug Report: Remove XGBoost < 3.1 constraint

[guillaume-vignal]

Describe the bug

When using XGBoost ≥ 3.1, the function
shap.explainers.Tree.supports_model_with_masker(model, None)
returns False, whereas it returned True with earlier versions of XGBoost.

This happens because learner_model_param["base_score"] in XGBTreeModelLoader.__init__ is now a list instead of a float, causing a type mismatch in the SHAP XGBoost model loader.

As a result, Shapash currently enforces a constraint on xgboost < 3.1, which prevents compatibility with recent versions.

---

Minimal reproducible example

import numpy as np
import pandas as pd
import xgboost as xgb
import shap

from shapash.backend.shap_backend import ShapBackend  # optional, same issue occurs with shap directly

model_list = [
    xgb.XGBRegressor(n_estimators=1),
    xgb.XGBClassifier(n_estimators=1),
]

df = pd.DataFrame(range(0, 21), columns=["id"])
df["y"] = df["id"].apply(lambda x: 1 if x < 10 else 0)
df["x1"] = np.random.randint(1, 123, df.shape[0])
df["x2"] = np.random.randint(1, 3, df.shape[0])
df = df.set_index("id")
x_df = df[["x1", "x2"]]
y_df = df["y"].to_frame()

model = xgb.XGBRegressor(n_estimators=1)
model.fit(x_df, y_df)

print(shap.explainers.Tree.supports_model_with_masker(model, None))
# Expected: True
# Actual (with xgboost>=3.1): False

Package	Version
SHAP	0.49.1
XGBoost	3.1.0+
Python	3.10+
OS	Ubuntu 22.04

---

Additional context

The next SHAP release includes a fix for this issue, but it is available only for Python ≥ 3.11 (see shap/shap#4202).
It would therefore be helpful to remove the strict XGBoost <3.1 constraint in Shapash and ensure internal compatibility with both XGBoost 3.0 and newer versions.