cranelift: Disable i128 divs on fuzzgen

[afonso360]

Do not merge!

This is going to change the fuzzer input format and invalidate the pending fuzzer issues.

Disables i128 div's from being generated since the x64 backend isn't ready for them yet.

Fixes: #4756
Fixes: #4770
cc: @cfallin @bnjbvr

[bjorn3]

For reference cg_clif uses compiler-builtins intrinsics for 128bit multiplication and division.

[afonso360]

I think we already support i128 multiplications (at least the test case is enabled), but I'd still like to implement i128 div/rem and cleanup cg_clif's codegen!

[bjorn3]

Won't 128bit div and rem be rather large? Using compiler-builtins avoids code bloat in those cases.

[afonso360]

Well the plan was to do exactly that, lower them as libcalls, or are they not available in all platforms?

[bjorn3]

I see. I believe it is available on all platforms. However LLVM has changed the abi of 128bit compiler builtins on windows before and this may happen again in the future.

[jameysharp]

When we have the other issues from the current batch of fuzzing sorted out, I approve these changes. Alternatively, if anyone wants to put together an x64 implementation of i128 division, we could merge that immediately since it doesn't change the interpretation of the fuzzer input. I don't think that implementation has to produce good code; as long as it's correct, we can improve its quality later.