fix: do not remove memory blocks used as brillig input

[guipublic]

Description

Problem*

Resolves #7062

Summary*

The unused memory optimisation was not checking for brillig inputs

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]

Execution Memory Report

Program	Peak Memory
keccak256	74.69M
workspace	123.49M
regression_4709	316.01M
ram_blowup_regression	512.61M
rollup-root	204.21M
rollup-merge	204.21M
rollup-block-root	204.21M
rollup-block-merge	204.21M
rollup-base-public	204.21M
rollup-base-private	204.21M
private-kernel-tail	168.36M
private-kernel-reset	168.39M
private-kernel-inner	168.36M

[github-actions]

Compilation Memory Report

Program	Peak Memory
keccak256	77.73M
workspace	123.85M
regression_4709	424.10M
ram_blowup_regression	1.58G
rollup-root	204.21M
rollup-merge	204.21M
rollup-block-root-single-tx	204.21M
rollup-block-root-empty	204.21M
rollup-block-root	204.21M
rollup-block-merge	204.21M
rollup-base-public	204.21M
rollup-base-private	204.21M
private-kernel-tail	168.36M
private-kernel-reset	168.39M
private-kernel-inner	168.36M

[TomAFrench]

Strange that we would be generating a memory block at all in this case. Can you open an issue to see if we can get it so that we're just passing in witnesses to this brillig function?

[vezenovm]

Strange that we would be generating a memory block at all in this case.

Anytime we initialize an array in ACIR we call a MemoryInit opcode.

noir/compiler/noirc_evaluator/src/acir/acir_variable.rs

Line 2020 in 9471e28

self.acir_ir.push_opcode(Opcode::MemoryInit {

We would probably need some way to lazily initialize arrays. Although in this test, that memory block is still used after the unconstrained call.