Function dot_i8_i32 

pub fn dot_i8_i32(
    left: &[u8],
    right: &[u8],
    left_zero_point: i8,
    right_zero_point: i8,
    bias: i32,
) -> Result<i32, IntegerError>

Compute an exact signed INT8 dot product with INT32 accumulation.

left and right contain the tensor’s two’s-complement storage bytes. Zero points are subtracted before multiplication, and bias initializes the accumulator. Wrapping is never used: a result outside INT32 is reported because TOSA classifies such an input as unpredictable rather than defining modular arithmetic.