feat(optimization): Redirect array gets to earlier arrays if possible

[jfecher]

Description

Problem*

Resolves comment #5772 (comment)

Summary*

This PR redirects array_gets to work on earlier arrays if they can (currently only block parameters.
This can potentially optimize code by allowing more array gets to be deduplicated but may also prevent optimizing array_sets to be mutable if older arrays are now being used more.

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 Brillig bytecode sizes

Generated at commit: fe0010828b76c355b3ded81f304f601a2530995e, compared to commit: 8dec84793d200dcb524aa5c397d0a84d38974e7e

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
brillig_arrays	-41 ✅	-36.61%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
brillig_arrays	71 (-41)	-36.61%

[github-actions]

Changes to circuit sizes

Generated at commit: fe0010828b76c355b3ded81f304f601a2530995e, compared to commit: 8dec84793d200dcb524aa5c397d0a84d38974e7e

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
conditional_1	-35 ✅	-0.76%	-125 ✅	-0.97%
array_if_cond_simple	-71 ✅	-63.96%	-312 ✅	-9.96%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
conditional_1	4,590 (-35)	-0.76%	12,725 (-125)	-0.97%
array_if_cond_simple	40 (-71)	-63.96%	2,822 (-312)	-9.96%

[TomAFrench]

Is this ready for review?

[jfecher]

Maybe. I'm still not sure if this optimization is worth it. Since it'd only have an effect on small programs I think due to the constant limit of how many instructions it can look past to. It also may lead to worse performance if that block parameter had a array_set_mut to it previously but can't anymore since now more array gets to it are inserted.

[jfecher]

Closing this PR, it should be redundant now that we have instruction deduplication in general