[WIP] Sparse Tensor Support

[andrewCodeDev]

The backend is there, but no functions currently use or work on sparse tensors. There are a few caveats:

1. We will not support full sparse tensor algebra. Adding two sparse tensors together with randomly initialized entries will almost always increase the density of the result tensor. Many operations are like this, and suddenly you're paying for a pseudo-sparse tensor that's actually more expensive than the dense variant. This would require heuristics and promotions.

2. Most frameworks use sparse tensors as a dense operation optimization. Underlying library support is dodgy, but for good reasons.

I'm leaning towards offering it as a separate class (like LogicalTensor) and only exposing the operations we need to support.