fix: optimizer to keep track of changing opcode locations

[aakoshh]

Description

Problem*

Resolves the issue found trying to merge AztecProtocol/aztec-packages#10483
Mirroring the commits added to that PR.

Summary*

Unlike transform_internal, optimize_internal did not take an acir_opcode_positions which would have allowed it to be chained. This caused the second iteration of the loop to reset acir_opcode_positions to the default values, which indicated no change when the AcirTransformationMap was derived from it. The consequence was that the assert_messages and debug information did not align with the new opcode locations, which changed in the first iteration.

Also changed test_cmd to apply the nargo::ops::transform optimisations, so we get to see any problems introduced there come up in tests. This actually brought out this regression in vecdequeue test, where an unexpected assertion message was returned.

Added an integration test to provoke the error fixed by the PR. In order for the ACIR not to eliminate all operations due to them working with constants known in the test code, I am calling it through the fuzzer, which makes the test look silly. To make it work required an update to the fuzz tester to anticipate tests that expect failures.

Additional Context

The way to debug this on aztec-packages went like this.
First build everything with ./bootstrap.sh, then in one terminal start an oracle:

export LOG_LEVEL=debug
cd aztec-packages/yarn-project/txe
yarn start

In another terminal run a test (one of the failing ones):

export NARGO=$(pwd)/aztec-packages/noir/noir-repo/target/release/nargo 
cd aztec-packages/noir-projects/noir-contracts
$NARGO test --silence-warnings --oracle-resolver http://localhost:8080 --package nft_contract transfer_to_public_failure_not_an_owner 

Then change something in the Noir codebase and recompile:

cd aztec-packages/noir 
./bootstrap.sh

With const MAX_OPTIMIZER_PASSES: usize = 1; the test passed, otherwise failed. The LOG_LEVEL=debug showed the index of the failing opcode. Curiously it was the same in both tests (the oracle is expected to fail, the test verifies it fails with the right message).

One of the things I tried was to print the ACIR opcodes with, with a change in the code so it prints the optimised version, after transformations:

$NARGO compile --package nft_contract --print-acir 

Comparing the ACIR indicated that 1 or 3 passes doesn't make any difference. Since it was also the same opcode, I started to suspect that it is actually the assertion message that gets lost.

With this fix the contract tests pass.

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]

Peak Memory Sample

Program	Peak Memory
keccak256	78.64M
workspace	121.88M
regression_4709	294.72M
ram_blowup_regression	2.44G
private-kernel-tail	208.79M
private-kernel-reset	862.88M
private-kernel-inner	307.69M
parity-root	174.43M

[github-actions]

Compilation Sample

Program	Compilation Time	%
sha256_regression	0m1.453s	-9%
regression_4709	0m1.480s	-4%
ram_blowup_regression	0m16.897s	-2%
private-kernel-tail	0m1.508s	1%
private-kernel-reset	0m9.642s	5%
private-kernel-inner	0m2.788s	17%
parity-root	0m1.117s	-5%
noir-contracts	2m51.576s	8%

[asterite]

Looks good!