Move load splat and load extend ops into the list of SIMD memory
opcodes, since they similarly take an i32 and an memarg. This fixes the
OpcodeLength calculation in function-body-decoder-impl.h.

And in turn, fixes the mjsunit test code that the fuzzer generates. See
the regress-1055692.js file for the weird S8x16LoadSplat followed by 2
kExprUnreachable, where the kExprUnreachable really is a memarg
{0x0, 0x0}. This bug was caught by the fuzzer, and that was the
generated test (with small fixes to add kExprDrop), so leaving it as it
is.

Bug: chromium:1055692
Change-Id: I743b6beb82350b5fea22c8dd10b546a02741cfed
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2071401Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66439}