feat: skip reading values immediately after it being written into an array

[TomAFrench]

Description

Problem*

Resolves

Summary*

I noticed a couple of cases where we were writing into an array and then immediately reading the same index from it. This seems like something that we should be able to unwind.

Draft as I want to do more thinking on how EnableSideEffects can interact with this.

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: b35f26b7bd67fc6202b85b6b5923e821899862dd, compared to commit: 299703cf4b87a84257f48f059eb58135ad36265d

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
array_if_cond_simple	-5 ✅	-4.31%	-12 ✅	-0.38%
slice_dynamic_index	-18 ✅	-1.48%	-35 ✅	-0.54%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
conditional_1	8,519 (-26)	-0.30%	38,799 (0)	0.00%
hashmap	209,812 (-84)	-0.04%	397,715 (-164)	-0.04%
regression_5252	81,672 (-114)	-0.14%	106,562 (-229)	-0.21%
regression_mem_op_predicate	52 (-6)	-10.34%	3,552 (-11)	-0.31%
array_if_cond_simple	111 (-5)	-4.31%	3,110 (-12)	-0.38%
slice_dynamic_index	1,200 (-18)	-1.48%	6,412 (-35)	-0.54%

[jfecher]

I can't think of situations where the current predicate would change this either.

The main case of

let mut a = ...;
if foo {
    a[0] = 1;
}
.. use a ...

Shouldn't apply here since (after removing stores/loads) we should see the later uses of a referring to the result of an IfElse instruction rather than an array set.

[jfecher]

Can we get this merged?

The optimization seems good to me.