[BUG] Prevent assignment from element-wise ops on same input

[cliffburdick]

Currently if a user has an input tensor that writes to the same output tensor this is fine as long as the element-wise operation doesn't change index positions. All scalar operators follow this pattern. However, for operators that change the output index, writing to the same buffer causes a race condition:

(a = fftshift1D(a)).run();

We need to do one of two things:

1. Raise an error when this situation occurs
2. Async-allocate memory to temporarily do the transform and make a copy afterwards.

2 is preferred because the user would have to do this anyways if they were to implement it themselves.

[luitjens]

I think we should start with 2. Then opportunistically we can detect cases where we don't need the temporary. We should take this approach to any operator in MatX that can cause race conditions in indexing. We should probably add a property to operators to specify if in-place is safe or not.