Refactor fusing (#1386)

This commit introduces handling graph fusion by encapsulating fusion metadata in the graph. This ensures that after access to the graph that modify the graph, a validation check is performed to verify that the fusion information remains consistent and that no modifications have introduced inconsistencies. Additionally, the fusion-related logic has been refactored into a new class called GraphFuser, which takes the graph along with its fusion metadata and creates a new graph where fused operations are represented as single nodes.

---------

Co-authored-by: reuvenp <reuvenp@altair-semi.com>

Author: reuvenperetz

.github/workflows/run_keras_tests.yml

@@ -27,11 +27,10 @@ jobs:
           pip install tensorflow==${{ inputs.tf-version }} sony-custom-layers 
           pip install pytest pytest-mock
           pip check
-      - name: Run unittests
-        run: |
-          python -m unittest discover tests/keras_tests -v
-
       - name: Run pytest
         run: |
           pytest tests_pytest/keras_tests
+      - name: Run unittests
+        run: |
+          python -m unittest discover tests/keras_tests -v
 

.github/workflows/run_pytorch_tests.yml

@@ -33,10 +33,11 @@ jobs:
           pip install torch==${{ inputs.torch-version }} torchvision onnx onnxruntime "onnxruntime-extensions<0.14"
           pip install pytest pytest-mock
           pip check
-      - name: Run unittests
-        run: |
-          python -m unittest discover tests/pytorch_tests -v
       - name: Run pytest
         run: |
           pytest tests_pytest/pytorch_tests
+      - name: Run unittests
+        run: |
+          python -m unittest discover tests/pytorch_tests -v
+