Refactor: removing scaling nodes

Added MulPTQ Operator and ScalingMeta for PTQ Module

This MR introduces the Scaling operator, which essentially implements a MUL operator within a MetaOperator. This operator is required for the PTQ (Post-Training Quantization) module.

Additionally, the Quantizer operator has been added to replace the now-deprecated ScalingNodes. The Quantizer acts as a meta-operator to handle scaling operations in the PTQ and removing the Scaling Node Operator (which is not compatible with ONNX).

Caution: This Merge Request aims to eventually replace the existing Scaling operator (which is not ONNX-compatible). As a result, the current operator is now considered deprecated.

For more details, please refer to the associated issues: #43 (closed)

aidge_quantization/unit_tests/test_ptq.py

@@ -17,7 +17,7 @@ from aidge_core import Log, Level
 NB_SAMPLES      = 1000               # max : 1000
 SAMPLE_SHAPE    = (1, 1, 28, 28)
 MODEL_NAME      = 'MiniResNet.onnx'  # 'ConvNet.onnx'
-ACCURACIES      = (95.4, 94.5)       # (97.9, 97.7)
+ACCURACIES      = (95.4, 94.4)       # (97.9, 97.7)
 NB_BITS         = 4
 
 # --------------------------------------------------------------
@@ -91,12 +91,13 @@ class test_ptq(unittest.TestCase):
 
         # quantize the model
 
+        
         aidge_quantization.quantize_network(
             self.model, 
             NB_BITS, 
             tensors, 
             clipping_mode=aidge_quantization.Clipping.MSE, 
-            apply_rounding=True, 
+            no_quantization=False,
             optimize_signs=True, 
             single_shift=False
         )