Function dequantize_f32 

pub fn dequantize_f32(
    data: &[u8],
    scales_f32: &[f32],
    group_len: usize,
    bits: u8,
    tensor_len: usize,
    frame_count: usize,
    out: &mut Vec<f32>,
)

Dequantize packed codes using f32 scales, writing f32 values.

Iterates by frame then by group to avoid per-value modulo/division and caches the f32 scale per group.

For 8-bit data, reads bytes directly without bit accumulation.