Feeding only output of last cell of LSTM to the FCL

[Jsevillamol]

Currently, the LSTM part of the CNN-RNN model is unrolled always 8 times, and its output concatenated and fed to a Fully Connected Layer (FCL).

We would like to only pass to the FCL the output of the last LSTM cell, to allow inputs of variable length. This would also potentially enable us to create an ensemble model, analyzing data of the same event but from different telescopes (we would need to standardize the image size with padding for this).