We should see if we can eliminate our null-checks for GC operations using our signal handling mechanism.

We currently do something similar to this for call_indirect, where the function we're calling will have InstanceReg = null when the function pointer is null. So we mark the instruction to load the HeapReg from InstanceReg as a potentially trapping instruction [1].

I believe we can just do a similar thing for most of our GC instructions. For all of them we need to emit the trap metadata immediately before the first load that would be done on a potentially null pointer.

[1] https://searchfox.org/mozilla-central/rev/99f83e78415867726a5076a525e4a60840f3be33/js/src/jit/MacroAssembler.cpp#4436

https://hg.mozilla.org/mozilla-central/rev/e4507c5f89e2