Bidirectional LSTM?

I'm in the process of porting some Tensorflow models over to Candle and I have a [bidirectional](https://github.com/keras-team/keras/blob/v2.14.0/keras/layers/rnn/bidirectional.py#L36-L533) LSTM layer in my model. Is this something that Candle can/will support in the future?

```python
tf.keras.layers.Bidirectional(
tf.keras.layers.LSTM(
self.parameters.lstm_units,
return_sequences=True)
)
```

Start with the linked Keras Bidirectional implementation and the shown LSTM configuration, then inspect Candle's current recurrent-layer entry points. Define what equivalent forward and reverse processing with return_sequences=True should produce; done requires a decided support path and validation against the TensorFlow model behavior.

Written by the indexing model from the issue text.