Function dequantize_affine 

pub fn dequantize_affine(data: &[i8], zp: i32, scale: f32) -> Vec<f32>

Convert raw INT8 values to f32 using affine dequantization.

RKNN models quantize weights and activations to INT8 during conversion. Each tensor has a zero-point (zp) and scale stored in its TensorAttr. This function reverses the quantization:

f32_value = (raw_i8 - zero_point) * scale

Example

use rknn_runtime::dequantize_affine;

let raw = vec![10i8, 20, 30];
let zp = 5;
let scale = 0.1;
let result = dequantize_affine(&raw, zp, scale);
assert_eq!(result, vec![0.5, 1.5, 2.5]);