fix: use range-constrained var from unsigned_overflow_check

[asterite]

Description

Problem

Resolves #8556
Resolves #8557
Resolves #8558

Summary

Applied this suggestions from @guipublic :

I think the solution should be to use the 'predicated' variable (predicate_range) in check_unsigned_overflow() as the result of the binary instruction. That should help for other things as well (e.g avoiding predicates in division by constant).
(The alternative is to use predicate when doing the truncation)

Additional Context

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.

[github-actions]

Changes to circuit sizes

Generated at commit: cd01741ab491d2f7eb233a53612e2c0b7bf856cb, compared to commit: 4972733638fa117f84848a3d923df1724bd84821

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
conditional_1	-55 ✅	-2.05%	-59 ✅	-0.73%
regression_8329	-3 ✅	-7.69%	-3 ✅	-5.45%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
slices	670 (+2)	+0.30%	3,769 (+2)	+0.05%
regression_3607	37 (0)	0.00%	2,800 (+1)	+0.04%
hashmap	33,473 (+4)	+0.01%	94,508 (+4)	+0.00%
slice_dynamic_index	812 (0)	0.00%	6,464 (-1)	-0.02%
conditional_1	2,633 (-55)	-2.05%	7,987 (-59)	-0.73%
regression_8329	36 (-3)	-7.69%	52 (-3)	-5.45%

[asterite]

@guipublic Is this what you had in mind?

If so, I wonder if the variable returned from range_constrain_var should be used in more places. Right now it's ignored in most cases.

[asterite]

Closed in favor of #8583
/