Ok so I:

1. Renamed {_values,_indices,_nnz} to {values,indices,nnz} (no underscore) in C++ and expose those in Python
2. Created wrappers for {_values,_indices,_nnz} on torch/tensor.py to preserve BC in Python
3. Rewrote C++ code to use .values() instead of .indices()

I think this should make everyone happy @zou3519 @ssnl @weiyangfb?

Renamed {_values,_indices,_nnz} to {values,indices,nnz} (no underscore) in C++ and expose those in Python

No, please don't do this. I'll quote the current sparse documentation:

Second, some operators will produce different values depending on whether or not they are coalesced or not (e.g., torch.sparse.FloatTensor._values() and torch.sparse.FloatTensor._indices(), as well as torch.Tensor._sparse_mask()). These operators are prefixed by an underscore to indicate that they reveal internal implementation details and should be used with care, since code that works with coalesced sparse tensors may not work with uncoalesced sparse tensors; generally speaking, it is safest to explicitly coalesce before working with these operators.

https://pytorch.org/docs/stable/sparse.html

@goldsborough Is your goal to unconditionally remove all underscore-prefixed methods from Tensor, or are you OK with exceptions? _indices and _values is something of a live wire in the current sparse tensor API, and can't really be adequately resolved without some careful documentation and addition of new methods. I actually do think that something like _indices or raw_indices should be in the steady state sparse API, but they should be clear that they have to be used with care.

@ezyang I'm ok with exceptions, I just thought we might as well fix the API since public underscore fields seem wrong either way. But maybe that's a discussion for another PR. I'll leave them be in this one :)

Test failures are real...