Extra allocation in sparse array benchmark

[gbaraldi]

run(BaseBenchmarks.SUITE[["sparse", "matmul", ("A_mul_B!", "sparse 4000x40, dense 40x40 -> dense 4000x40")]])
got an extra allocation in #48977 @ViralBShah @Wimmerer

[rayegun]

Matmul did?

That is... Highly unexpected

[gbaraldi]

The performance is the same, but it went from 0 to 1 allocation.

[KristofferC]

Is it anything to care about?

[gbaraldi]

Extra allocations where we had 0 before are kind of annoying. I think it deserves a look, if it turns out it's unavoidable then fine, but I wouldn't want it going through without a look at least.

[vtjnash]

Allocations are not really a problem. And should be opened on SparseArrays repo if it is