fix: prevent OutputStackOverflow in intrinsic i32::overflowing_sub

[mooori]

Closes #1163

On top of #1160 to make use of intrinsic testing.

Problem

compiler/codegen/masm/intrinsics/i32.masm

Line 92 in 30544eb

drop push.MAX # [i32::MAX, a]

The drop would cause stack depth to shrink below the minimum of 16, so the processor implicitly adds a stack slot by adjusting the bottom of the stack [ref].

So after drop push.MAX the stack depth is 17 while in the intrinsic it's assumed to be 16. The extra slot isn't cleaned up, which results in OutputStackOverflow.

Footgun?

MASM authors should probably be aware of this, still it can be a footgun. Could this be:

• turned into a warning/error produced by a linter?
• prevented by the processor by trying to recognize such scenarios and remove the extra slot if possible?
  • though this might not always work reliably and would break MASM which builds on this behavior

[greenhat]

Looking good!

[bitwalker]

This should not have been merged - this behavior in the processor with regard to the operand stack is known and expected, and is why we emit calls to core::sys::truncate_stack at the end of every program entrypoint - which is what should have been done here.

Let's revert this and adjust the test code instead to call that procedure like we do during codegen