In current BitwiseSmi bytecodes the code to do Smi operation is inside a
loop. This CL fast path the Smi operation by peeling the first Smi check
out of the loop, and avoid Smi->Int->Smi conversion where possible.

Drive-by fix: Add CSA_DCHECK in Smi shift to avoid unexpected use.

Bug: v8:12442
Change-Id: I1adce560fb22a4409337e2958779eccf9197e4ff
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3328784Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Commit-Queue: Hao A Xu <hao.a.xu@intel.com>
Cr-Commit-Position: refs/heads/main@{#78764}