fix(random): random_bit_list should set exact length

[iamquang95]

Should fix this: #317

[iamquang95]

@claude

[claude]

Claude finished @iamquang95's task in 1m 17s —— View job

---

Code Review

• Read changed file
• Check Go reference implementation
• Analyze correctness of the fix
• Post review feedback

---

Summary

The fix correctly addresses a real bug where random_bit_list could set fewer than length bits due to duplicate indices. The original code looped length times, but could land on the same bit multiple times via rng.gen::<usize>() % 256, giving you a bitlist with fewer set bits than requested.

---

Analysis

The root cause (original code):

// Old: duplicate bit_idx values possible → fewer than `length` bits actually set
for _ in 0..length {
    let bit_idx = rng.r#gen::<usize>() % 256;
    bytes[byte_idx] |= 1 << bit_offset;
}

The fix:

// New: index::sample guarantees `length` distinct indices
for bit_idx in index::sample(&mut rng, 256, length) { ... }

rand::seq::index::sample draws length unique indices from [0, 256) without replacement — exactly what's needed here. This is the correct, idiomatic fix.

---

Notes

1. Intentional divergence from Go — The Go reference (charon/testutil/random.go) has the same duplicate-index issue:

func RandomBitList(length int) bitfield.Bitlist {
    size := 256
    resp := bitfield.NewBitlist(uint64(size))
    for range length {
        resp.SetBitAt(uint64(rand.Intn(size)), true)  // duplicates possible
    }
    return resp
}

This PR makes Pluto's version stricter than the Go counterpart. Per AGENTS.md, the default is functional equivalence with Go, but since this is test utility code and the Go version's behavior appears to be a latent bug rather than an intentional design choice, the fix is justified. The overall test semantics (generating a bitlist with some bits set) are preserved.

2. The assert! guard is appropriate — Calling index::sample(rng, 256, n) with n > 256 would panic with a cryptic message from inside rand. The explicit assert! with a clear message is good practice.

3. Edge cases — length = 0 produces index::sample(&mut rng, 256, 0) which returns an empty iterator, leaving bytes all-zero. That's correct behavior.

---

Verdict

The fix is correct, clean, and idiomatic. No issues found. ✅