Apply _convert_to_tensor in exp implementation in NumPy/JAX backends.

Patched from d9deed0 .

Author: jburnim

tensorflow_probability/python/internal/backend/numpy/numpy_math.py

@@ -568,7 +568,7 @@ def _divide_no_nan(x, y, name=None):  # pylint: disable=unused-argument
 
 exp = utils.copy_docstring(
     'tf.math.exp',
-    lambda x, name=None: np.exp(x))
+    lambda x, name=None: np.exp(_convert_to_tensor(x)))
 
 expm1 = utils.copy_docstring(
     'tf.math.expm1',