Add RNN wrapper for Cells

[rxwei]

Based off of #100 by @tanmayb123. My push to that branch accidentally closed that PR, so I'm starting a new one.

• Add generic structure type RNN<Cell: RNNCell> that forms a recurrent layer from a cell. Thanks @tanmayb123 for starting this!
• Define a custom efficient derivative for RNN.call(_:).
• Make RNN structures (SimpleRNNCell, LSTMCell, RNN) conditionally conform to VectorNumeric and VectorNumeric for easier integration with more efficient optimizers in the future.
• Make RNN cell initializers take a random seed, passing them through Tensor.init(glorotUniform:seed:).
• Add a test for RNN using SimpleRNNCell.

Related to #52. Resolves #91.

[googlebot]

So there's good news and bad news.

👍 The good news is that everyone that needs to sign a CLA (the pull request submitter and all commit authors) have done so. Everything is all good there.

😕 The bad news is that it appears that one or more commits were authored or co-authored by someone other than the pull request submitter. We need to confirm that all authors are ok with their commits being contributed to this project. Please have them confirm that here in the pull request.

Note to project maintainer: This is a terminal state, meaning the cla/google commit status will not change from this state. It's up to you to confirm consent of all the commit author(s), set the cla label to yes (if enabled on your project), and then merge this pull request when appropriate.

ℹ️ Googlers: Go here for more info.

[rxwei]

For the record, CLA passed in #100.

[rxwei]

Note: Regarding this array reversal, while I could've zero-initialized an array of timeStepCount tensors and modified them in reverse order, it would be less efficient because of the cost of heap-allocating timeStepCount extra tensors.

[tanmayb123]

I think CI isn't working.

[rxwei]

Yup, still investigating. Locally all tests are passing.