Get the initialization neural network working

[eigenvivek]

Nalini's advice:

1. Check the training loss
2. Make sure the network can learn a single example
3. Use batchnorm
4. Use residual connections
5. Try a deeper network
6. Try to predict just a few parameters (eg, just project through an axis-aligned slice)

[eigenvivek]

NN works perfectly well with MSELoss. The issue is defining a loss function with the DRR module. This is because the DRR module makes the geometrics base parameters... that is, I can't pass parameters from another layer without losing the gradient

Fix will require a change to the API to input torch.tensors (will implicitly fix #30, but will break all other examples).

[eigenvivek]

Or one can simply make the projection its own function for use as an intermediate layer :)

[eigenvivek]

Batched DRR rendering was completely broken (#107), but fixed now!