fix: zero out input to to_radix calls if inactive

[TomAFrench]

Description

Problem*

Resolves

Summary*

In situations where a to_le_bytes call is in an inactive if-statement, the requirement for the produced byte array to represent the input field is not disabled. This can result in situations such as below where we want to decompose a value into a number of bytes which is dependent on the inputs to the circuit.

https://github.com/porco-rosso-j/safe-recovery-noir/blob/b80a1fd49d370bd095827318b378c8fdef2e2d9b/circuits/social/src/root.nr#L11-L22

All branches of this snippet are limited by the fact that the first will fail on index with values greater than 1.

fn main(x: Field, cond: bool) {
    if cond {
        let bad_byte_array = x.to_le_bytes(1);
        assert_eq(bad_byte_array.len(), 1);
    }
}

This compiles down to

acir fn main f0 {
  b0(v0: Field, v1: u1):
    enable_side_effects v1
    v31, v32 = call to_le_radix(v0, u32 2⁸, u32 1)
    v34 = cast v1 as Field
    v35 = mul v31, v34
    constrain v35 == v34
    enable_side_effects u1 1
    return 
}

The to_le_radix will revert for all v0 >= 8 no matter the value of v1.

I've modified the flatten_cfg pass such that we multiply in the side effects variable to zero out the input should the instruction be deactivated.

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [Exceptional Case] 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]

Changes to circuit sizes

Generated at commit: e5e72ad48198daca06d56ba4c4be1f39211b7d10, compared to commit: d1740dda2d54fb3cff3df06eb08c10ae6d9fd865

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
to_bytes_integration	+508 ❌	+819.35%	+507 ❌	+429.66%
bit_shifts_runtime	+1,520 ❌	+193.14%	+1,517 ❌	+39.43%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
to_bytes_integration	570 (+508)	+819.35%	625 (+507)	+429.66%
bit_shifts_runtime	2,307 (+1,520)	+193.14%	5,364 (+1,517)	+39.43%
to_le_bytes	66 (+4)	+6.45%	153 (+3)	+2.00%
u128	891 (+57)	+6.83%	4,727 (+72)	+1.55%
merkle_insert	1,988 (+15)	+0.76%	29,032 (0)	0.00%
operator_overloading	596 (+251)	+72.75%	8,214 (0)	0.00%

[vezenovm]

Should we also change the specification that ToBits and ToRadix have side effects under instruction.rs?

[TomAFrench]

Should we also change the specification that ToBits and ToRadix have side effects under instruction.rs?

Good point, yeah we should add that.