The interp stubs for exports that pass v128 must handle the v128 ending up on the stack. My feeble tests for v128 did not hit this case (they only hit the register case and I neglected to make the obvious connection that the stack case would have to be handled too) but as I was noodling around with the x86 port I hit this pretty quick, and it will be needed for x64 too.

We can't MOZ_CRASH for the v128 cases of the stack copies in
SetupABIArguments and StackCopy. In the former case we can emit a
breakpoint instruction because the code should never be executed. In
the latter case we just copy the data.

Adds test cases that triggered the MOZ_CRASHes, and other test cases
that are related but left out of the test cases when SIMD for baseline
first landed.

https://hg.mozilla.org/mozilla-central/rev/9d54a4e4532c