fix: Prevent hoisting binary instructions which can overflow

[asterite]

Description

Problem

Resolves #6665

Summary

We are hoisting any instruction that doesn't have side effects. But sub (and I guess any binary operation?) can lead to underflow/overflow so we can't always hoist it: if that sub can only be done after we check that a value is not zero, doing it before that will lead to that panic.

In this PR we now consider those binary ops as having side effects (the definition in that method is "can fail or interacts with memory".

Additional Context

A code that does sub and underflows exists in Aztec-Packages. I tried to reduce it to something that we can use as a test but couldn't... in any case the SSA program alone should be a good test for it, regardless of how that SSA got to that state.

Documentation

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[TomAFrench]

I guess we can consider Add, Sub, Mul and Div as having side effects? Alternatively we can drop the has_side_effects name and just use can_be_hoisted.

Pending #6623 where we can make SSA aware of when we need to do overflow checks then I think we have to consider them as such (minus cases such as division by a constant, etc.) I'd be up for a can_be_hoisted helper function here for the time being.

[github-actions]

Changes to number of Brillig opcodes executed

Generated at commit: 7eef6179bf958e9dc9115fe0073c4e7f6a78b0e8, compared to commit: 6acef6d795cd74dee4a21e82c5a912d58b40b06c

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
nested_array_dynamic	+105 ❌	+3.38%
break_and_continue	-3 ✅	-2.59%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
nested_array_dynamic	3,214 (+105)	+3.38%
aes128_encrypt	4,499 (+48)	+1.08%
tuple_inputs	633 (+1)	+0.16%
slice_dynamic_index	4,304 (-1)	-0.02%
break_and_continue	113 (-3)	-2.59%

[github-actions]

Changes to Brillig bytecode sizes

Generated at commit: 7eef6179bf958e9dc9115fe0073c4e7f6a78b0e8, compared to commit: 6acef6d795cd74dee4a21e82c5a912d58b40b06c

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
regression_3607	+4 ❌	+8.33%
nested_array_dynamic	+164 ❌	+8.27%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
regression_3607	52 (+4)	+8.33%
nested_array_dynamic	2,148 (+164)	+8.27%
break_and_continue	64 (+4)	+6.67%
aes128_encrypt	520 (+8)	+1.56%
nested_array_in_slice	1,105 (+8)	+0.73%
tuple_inputs	366 (+2)	+0.55%
regression	941 (+4)	+0.43%
slice_dynamic_index	2,533 (+1)	+0.04%