Keras LSTM-RNN layer

Please make sure that this is a feature request. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:feature_template

System information

• TensorFlow version (you are using): 2.3.1
• Are you willing to contribute it (Yes/No): Yes, when able and available

Describe the feature and the current behavior/state.
There is a high-level API on Keras to LSTM layers on top of RNN that allows getting LSTM output as simple as this:

lstm_module = LSTMModule(5)   
lstm_input = tf.constant([[0.1, 0.2], [0.3, 0.4]], shape=[1, 2, 2])
lstm_output = lstm_module(lstm_input)

A definition of the LSTM Layer with Model Subclassing API from Tensorflow:

class LSTMModule(tf.keras.layers.Layer):

    def __init__(self, lstm_dims):
        super().__init__()
        self.lstm_dims = lstm_dims
        self.lstm = LSTM(lstm_dims, return_sequences=True, return_state=True)

    def call(self, inputs):
        # Forward pass
        ini_hidden_state = tf.zeros(shape=[1, self.lstm_dims]), tf.zeros(shape=[1, self.lstm_dims])
        return self.get_lstm_output(self.lstm, inputs, ini_hidden_state)

    @staticmethod
    def get_lstm_output(lstm_model, input_sequence, initial_state):
        output = lstm_model(input_sequence, initial_state=initial_state)
        hidden_states, hidden_state, cell_state = output[0], output[1], output[2]
        return hidden_states, hidden_state, cell_state

Will this change the current api? How?
This will add a new feature to tensorflow-framework module.

Who will benefit with this feature?
Anyone that requires deep learning to solve sequence classification and prediction problems and everyone who is already familiar with Keras.

Any Other info.
This feature comes from #270

未指定文件或测试。首先审查 issue #270，以及现有的、与 Keras 风格 LSTM/RNN 层请求相关的 tensorflow-framework API。完成的标准是确定所请求的高级 LSTM API 是否属于此仓库，并定义其实现范围。