PyTorch with `xsmm.zero` left-over before input online packing

[rengolin]

The PyTorch models we have in the benchmarks get a left-over xsmm.zero for the entire (unpacked) input in addition to the one inside the loop (that gets converted to beta=0 on the GEMM). The IR gets different at the tensor level, before bufferizaiton, so it could be a pack propagation / canonicalization issue?

@nhasabni @chelini