Function ste_forward 

pub fn ste_forward(
    input: &Tensor,
    scale: f32,
    min_val: f32,
    max_val: f32,
) -> Result<Tensor>

Apply STE forward pass (quantize then dequantize).

This creates a “fake quantized” tensor that:

• Has values as if they were quantized and dequantized
• Can receive gradients normally during backprop

Arguments

• input - Input tensor to quantize
• scale - Scale factor for quantization
• min_val - Minimum quantized value (e.g., -1 for ternary)
• max_val - Maximum quantized value (e.g., +1 for ternary)

Errors

Returns error if tensor operations fail.