Handle vector RHS in sparse solver routine

[tvercaut]

As pointed out by @tangzhenjie in #50, using our sparse solver in the case of a vector RHS fails in the backward pass.

Unsqueezing x in SparseGenericSolve::backward and the like should solve the problem

torchsparsegradutils/torchsparsegradutils/sparse_solve.py

Line 265 in 0004528

A, x = ctx.saved_tensors

        # Unsqueeze, if necessary
        is_vector = (x.ndimension() == 1)
        if is_vector:
            x = x.unsqueeze(-1)

Unit tests based on the following MRE should be added:

import torch
import torchsparsegradutils as tsgu

A = torch.randn(12, 12, requires_grad=True).to_sparse()
b = torch.randn(12, requires_grad=True)

x = tsgu.sparse_generic_solve(A, b)
loss = x.sum()
loss.backward()