cumsum!/cumprod! wrong when dim>N

[andreasnoack]

Discovered by @shashi:

julia> cumprod([2], 2)
1-element Array{Int64,1}:
 4

julia> cumprod([1 2; 3 4], 3)
2x2 Array{Int64,2}:
 1   4
 9  16

cc: @timholy

[sglyon]

I'm close to a fix here

[timholy]

Sorry for the bug. And thanks @spencerlyon2 for tackling it.

FWIW I don't see any reason this needs to use Base.Cartesian anymore (but whatever fix arrives should be applied).

[andreasnoack]

It might be worth to consider #14730 as part of the fix if someone decides to rewrite the functions anyway.

[sglyon]

I think the fix is a simple as adding this before any Cartesian magic happens:

if axis > N
    copy!(B, A)
    return B
end

[timholy]

That fix sounds rather sensible to me. And 👍 to #14730 as a more general approach.