Add Simulated Quantized Linear Operator for Debugging

[Fridah-nv]

Similar to the torch reference ops we have for attention and RoPE, create pure Python implementation for each quantized op to make it easier to debug our pipeline.