fix: support for conditional stores

[guipublic]

Description

Problem*

Resolves #2493

Summary*

Array Set with side-effect context are replaced by predicate*value+(1-predicate)array read
The index of array operation is also replaced by predicateindex.

This strategy does not work with empty slice since we would access index 0 and that would still fail with an empty slice, but I believe this case is degenerate enough to not support it.

Documentation

• This PR requires documentation updates when merged.

  • I will submit a noir-lang/docs PR.
  • I will request for and support Dev Rel's help in documenting this PR.

Additional Context

The PR requires ACVM PR 524 for the CI tests to pass.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[vezenovm]

This looks to be failing with the same bug I have here (#2446). I'm working on a fix as I think I may know the issue.

[vezenovm]

The failing test should pass now if you pull master.

[vezenovm]

Could we add a test like below to array_dynamic where idx is greater than the length of x?

fn regression_mem_op_predicate(mut x: [u32; 5], idx: Field) {
  // We should not hit out of bounds here as we have a predicate
  // that should not be hit
  if idx as u32 < 3 {    
    x[idx] = 10;
  }
  assert(x[4] == 111);
}

[vezenovm]

Just some comments suggestions mainly, but none are blocking for getting this out

[jfecher]

LGTM. We can merge once merge conflicts are fixed