The two instructions are fused into a single Uadalp instruction,
improving performance of quantized neural network operator
implementations such as XNNPACK.

Bug: v8:11546
Change-Id: Ic11b35d1e7758ee0b4ccfe8f592edc1aa798f6f2
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2939997Reviewed-by: Zhi An Ng <zhin@chromium.org>
Commit-Queue: Daan de Graaf <daagra@google.com>
Cr-Commit-Position: refs/heads/master@{#75102}