pub fn dot_bit4_batch(q: &[i8], rows: &[u8], d: usize, out: &mut [i32])Expand description
Scores one signed-byte query against contiguous packed four-bit rows.
Callers must pre-validate q.len() == d and rows.len() == d.div_ceil(2) * out.len() and d <= MAX_DOT_I8_DIMENSION. A zero dimension
fills out with zero. No scratch storage is allocated.