Add step==0 validation in WebGPU Slice operator
The CPU Slice implementation validates step != 0 (slice_helper.h:113)
but the WebGPU implementation was missing this check. A zero step
causes division-by-zero in the output dimension calculation
(ceil((end-start)/step)), which produces undefined behavior via
static_cast<int64_t>(inf).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
