Detect accumulation-like indexing

[mcabbott]

Motivated by #115, this tries to detect indexing of this pattern:

x = rand(Int8, 10) .+ 0; y = copy(x)
@tullio y[i] = y[i-1] + y[i]

i.e. writing in-place, to an array which appears on the right, with a shifted index. This i is very unsafe, worse than merely causing race conditions if multi-threaded, the expected result y == cumsum(x) depends on it being done sequentially.

It's possible this should be made an error. But without a shift, @tullio y[i] = y[i] + 1 seems like perfectly cromulent index notation for y .+= 1. And with a new array, of course @tullio z[i] := y[i+1] - y[i] is also fine, z == diff(y).

[mcabbott]

The second issue in #115 was about fixing indices on the LHS within shifts. Toy example which now works:

A = zeros(Int, 3,4);
for r in 2:4
    @tullio A[$r-1, c] = c + $r^2 
end
A == [c + r^2 for r in 2:4, c in 1:4]

Two maybe surprises are (1) that the index r isn't checked in any way, for r in 5:10 simply writes out of bounds. And (2) LoopVectorization gives the wrong answer:

using LoopVectorization
A = zeros(Int, 3, 4);
for r in 2:4
    @avx for c in 1:4
        A[r - 1, c] = c + r^2
    end
end
A != [c + r^2 for r in 2:4, c in 1:4]