`reshape` fails when expanding/reducing dims for one-element scalars/tensors

[mtsokol]

Hi @willow-ahrens,

The pydata/sparse#865 is almost ready - we only have a few edge cases for jl.reshape function missing.

ATM reshape doesn't cover cases where input is a scalar and reshape is meant to expand dims. Likewise, for one-element tensors when dropping dims (and hypothesis in Array API tests came across these cases).

Here's a script that reproduces it:

using Finch

T1 = swizzle(Tensor(Element(0, 0)))
T2 = swizzle(Tensor([0]), 1)

reshape(T1)
reshape(T1, 1)  # fails
reshape(T1, 1, 1)  # fails

reshape(T2, 1)
reshape(T2, 1, 1)
reshape(T2)  # fails