fix failing xgboost tests in nightly build by upgrading to latest shap, which is getting pinned by econml->responsibleai to older version

[imatiach-msft]

Seeing nightly build failures in interpret-community package for latest v2.0.0 xgboost models (just recently released to pypi). This error is already fixed in latest shap but when we install responsibleai package, the econml package downgrades shap because it is pinned to an older version. This PR undoes the pin temporarily in the builds to install latest shap which resolves the test failures.

Build failure link:

https://dev.azure.com/responsibleai/interpret-community/_build/results?buildId=16869&view=logs&j=7f33e5bd-7764-5d8a-ba2e-506e078b9c3f&t=9dfda548-ca11-5cc8-5a69-382ee7f1c50a

Build failure error:

        # shortcut using the C++ version of Tree SHAP in XGBoost, LightGBM, and CatBoost
        if self.feature_perturbation == "tree_path_dependent" and self.model.model_type != "internal" and self.data is None:
            model_output_vals = None
            phi = None
            if self.model.model_type == "xgboost":
                import xgboost
                if not isinstance(X, xgboost.core.DMatrix):
                    X = xgboost.DMatrix(X)
                if tree_limit == -1:
                    tree_limit = 0
                try:
>                   phi = self.model.original_model.predict(
                        X, ntree_limit=tree_limit, pred_contribs=True,
                        approx_contribs=approximate, validate_features=False
                    )
E                   TypeError: predict() got an unexpected keyword argument 'ntree_limit'

/usr/share/miniconda/envs/interp_community/lib/python3.8/site-packages/shap/explainers/_tree.py:332: TypeError