fix(ssa): Replace pop from 0-length slice with constraint and defaults

[aakoshh]

Description

Problem*

Resolves #9467
Replaces #9471

Summary*

If we see a v1, v2, v3 = call slice_pop_front(u32 0, v0) which is statically known to fail, we replace it with:

1. constrain 0 == <side-effect-variable>, "Index out of bounds" which will fails iff the pop was enabled, and
2. v3 = make_array [] which creates a default value for the result, so that we can still compile and execute the SSA, without executing the pop, which would fail, because it ignores the side effect variable.

Additional Context

In #9471 my approach was to add the side effect variable as a predicate to the ACIR memory operation, turning it into a no-op if the branch was not enabled. However, Barretenberg doesn't handle memory operation predicates, so this was a no go.

In this PR we instead look for a constant zero length, and replace it with a constraint that fails when pop would have failed, but does not fail if the side effects are disabled. In both cases we can remove the pop.

We can only handle the case when we statically know that the slice is empty. If the length is a variable, and it depends on a condition, then it will still fail at runtime if it's empty, even if the side effects are disabled. In my testing I was able to create a situation where the length was a variable by merging slices of unequal length, but in these cases the storage was not empty, and it successfully popped even if the length was 0. For example let s: [u32] = if b { &[] } else { &[10] }) results in v0 = b * 1; v1 = make_array [b * 10], and despite length being 0, popping from [0] caused no "Index out of bounds".

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]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Test Suite Duration'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: 8d4373e	Previous: ee2ac1a	Ratio
test_report_zkpassport_noir-ecdsa_	3 s	2 s	1.50
test_report_zkpassport_noir_rsa_	1 s	0 s	+∞

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

[jfecher]

We can only handle the case when we statically know that the slice is empty. If the length is a variable, and it depends on a condition, then it will still fail at runtime if it's empty, even if the side effects are disabled.

Can you make sure we have an issue for the runtime case?

[aakoshh]

67adec6 was required because of a fuzzing failure:

NOIR_AST_FUZZER_SEED=0x5624a20000100000 cargo test -q -p noir_ast_fuzzer_fuzz min_vs_full

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Execution Time'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: 8d4373e	Previous: ee2ac1a	Ratio
private-kernel-inner	0.019 s	0.014 s	1.36

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench