pub fn batch_dot_scores(
query: &[f32],
vectors: &[f32],
dim: usize,
scores: &mut [f32],
)Expand description
Batch dot-product scoring: f32 query vs N contiguous f32 unit-norm vectors.
For pre-normalized vectors (||v|| = 1), cosine = dot(q, v) / ||q||.
Skips per-vector norm computation — ~40% less work than batch_cosine_scores.