[quantization] Convert Graph to True Integer Representation After PTQ and Optimize Quantizers with Bitshift

This issue focuses on improving the efficiency of the graph after Post-Training Quantization (PTQ) by:

• Casting the graph from fake quantization (floating-point with scaling factors) to a true integer representation.
• Replacing quantizer nodes with bitshift operations when the "single shift approximation" feature is enabled.

Fixed by MR:!46 (merged) This is related to this issue: aidge_backend_cpu#35 (moved)