Add support for RNNs and LSTMs in ennuf models

Requires completion of #8 and #9 first.

Need to write representations of these such that we can define some model architecture that makes use of RNNs or LSTMs, and call some sort of
`.to_fortran()` method on it that generates Fortran corresponding to that structure, calling any components previously implemented in #8 and #9.

First review the work required by issues #8 and #9, since this issue depends on both. Then define how RNN and LSTM representations fit into a model architecture and how .to_fortran() should generate Fortran using the earlier components. Done means architectures using RNNs or LSTMs can be represented and translated to corresponding Fortran.

Written by the indexing model from the issue text.