fix: Do not carry over #[fold] to unconstrained functions during monomorphization

[aakoshh]

Description

Problem*

Resolves #10153

Summary*

Check whether the primary inline function attribute is compatible with the function we are creating; if not, use the default inline type.

Additional Context

During monomorphization when we call an ACIR function from an unconstrained one, it changes the runtime and creates an unconstrained version. During this the doppelgänger function variant inherited all the function attributes from the original, including ACIR-only attributes that the front end would have rejected if it saw them.

Another instance where this slipped in was the use function pointers: to be able to pass them on without knowing whether they will be called from constrained or unconstrained context, we create a both versions of them.

Here's the monomorphized AST showing the integration test no longer containing the #[fold] for the unconstrained version of foo:

fn main$f0() -> pub Field {
    let _f$l0 = (foo$f1, foo$f2);
    {
        bar$f3()
    }
}
#[fold]
fn foo$f1() -> Field {
    0
}
unconstrained fn foo$f2() -> Field {
    0
}
unconstrained fn bar$f3() -> Field {
    foo$f2()
}

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.

[TomAFrench]

LGTM after removing unnecessary Prover.toml

[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: bff82c7	Previous: cedc8ec	Ratio
test_report_zkpassport_noir-ecdsa_	3 s	2 s	1.50

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

CC: @TomAFrench

[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: bff82c7	Previous: cedc8ec	Ratio
rollup-block-root-single-tx	0.003 s	0.002 s	1.50

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

CC: @TomAFrench

[TomAFrench]

Hmm, that's an odd error.

[aakoshh]

Yes, apparently Lints disallow no_predicates on Brillig functions, but in fact they are required for inlining to get rid of some calls in these tests that must not survive until Brillig-gen

[aakoshh]

I'm checking why it wasn't just simplified out on its own.

[aakoshh]

The reason we need to keep the no_predicates is because derive_pedersen_generators cannot reach Brillig-gen, but they can only be simplified out if its arguments are constants. In the pedersen_check we have the following SSA:

//After Flattening (1) (step 22):
g0 = Field 340282366920938463463374607431768211456

brillig(inline) predicate_pure fn main f0 {
  ...
  b3():
    ...
    v28 = make_array [v1, v2] : [Field; 2]
    v30 = call f1(v28, u32 0) -> Field
    ...
    return
}
brillig(no_predicates) predicate_pure fn pedersen_hash_with_separator f1 {
  b0(v1: [Field; 2], v2: u32):
    v5 = make_array [Field 0, Field 0, Field 0, Field 0, Field 0, Field 0] : [(Field, Field); 3]
    v6 = allocate -> &mut [(Field, Field); 3]
    store v5 at v6
    v8 = make_array [Field 0, Field 0, u1 1, Field 0, Field 0, u1 1, Field 0, Field 0, u1 1] : [(Field, Field, u1); 3]
    v9 = allocate -> &mut [(Field, Field, u1); 3]
    store v8 at v9
    v25 = make_array b"DEFAULT_DOMAIN_SEPARATOR"
    inc_rc v25
    v27 = call derive_pedersen_generators(v25, v2) -> [(Field, Field, u1); 2]
    jmp b1(u32 0)
  ...
}

So, even though mem2reg, which comes next, does try to simplify this, it can't because v2 is a parameter to pedersen_hash_with_separator.

Once we get to the "Inlining" pass, it gets inlined into main and disappears:

After Inlining (2) (step 24):
g0 = Field 340282366920938463463374607431768211456

brillig(inline) predicate_pure fn main f0 {
  ...
  b3():
    ...
    v29 = make_array [v1, v2] : [Field; 2]
    v30 = make_array [Field 0, Field 0, Field 0, Field 0, Field 0, Field 0] : [(Field, Field); 3]
    v31 = allocate -> &mut [(Field, Field); 3]
    store v30 at v31
    v32 = make_array [Field 0, Field 0, u1 1, Field 0, Field 0, u1 1, Field 0, Field 0, u1 1] : [(Field, Field, u1); 3]
    v33 = allocate -> &mut [(Field, Field, u1); 3]
    store v32 at v33
    v49 = make_array b"DEFAULT_DOMAIN_SEPARATOR"
    inc_rc v49
    v50 = make_array [Field 3728882899078719075161482178784387565366481897740339799480980287259621149274, Field -9903063709032878667290627648209915537972247634463802596148419711785767431332, u1 0, Field 2393473289045184898987089634332637236754766663897650125720167164137088869378, Field -7135402912423807765050323395026152633898511180575289670895350565966806597339, u1 0] : [(Field, Field, u1); 2]
    jmp b4(u32 0)