Memory leak using tf.py_function in customized embedding layers with eager execution

[workingloong]

elasticdl/elasticdl/python/elasticdl/layers/embedding.py

Lines 176 to 178 in 020c928

	batch_embedding = tf.py_function(
	self.lookup_embedding, inp=[unique_ids], Tout=tf.float32
	)

The memory leak bug about tf.py_function has been submitted to TensorFlow issue 35010