Force numpy array coercion to 'object'

See NEP 0034: https://numpy.org/neps/nep-0034-infer-dtype-is-object.html

matminer/featurizers/structure/bonding.py

@@ -532,7 +532,7 @@ def fit(self, X, y=None):
             self
         """
         unpadded_bobs = [self.bag(s, return_baglens=True) for s in X]
-        bonds = [list(bob.keys()) for bob in unpadded_bobs]
+        bonds = np.asarray([list(bob.keys()) for bob in unpadded_bobs], dtype=object)
         bonds = np.unique(sum(bonds, []))
         baglens = [0] * len(bonds)