Function recall_scale 

pub fn recall_scale(
    target_recall: Option<f32>,
    base_ef: usize,
    base_oversample: u8,
) -> Result<(usize, u8), RerankError>

Scale ef_search and oversample up to meet a recall target.

target_recall must be in (0.0, 1.0]. Returns (ef, oversample) adjusted for the given target. When target_recall is None or already-default, returns the base values unchanged.

Formula (heuristic, not a guarantee):

• For r <= 0.80: identity — ef = base_ef, oversample = base_oversample.
• For r > 0.80: ramp scale from 1.0× at r=0.80 to 5.0× at r=1.00, linearly. scale = 1.0 + (r - 0.80) / 0.20 * 4.0, clamped to [1.0, 5.0].
• ef becomes max(base_ef, (base_ef as f32 * scale).ceil() as usize).
• oversample becomes max(base_oversample, (base_oversample as f32 * scale.sqrt()).ceil() as u8) — oversample grows sub-linearly because rerank cost is linear in oversample, while ef has a more favourable cost curve.