[Continue] Do not ignore the entire fused pattern if non-primary nodes are in the ignored scopes

[ljaljushkin]

Changes

Adjusted quantizer propagation common algo to only ignore the entire fused pattern if the primary node from the pattern ends up matching to an ignored scope.

Reason for changes

Improves logical consistency of the "ignored_scopes" and its effect on the algo result.

Related tickets

92302

Tests

updated references + new test cases in test_activation_ignored_scope

[ljaljushkin]

Some tests were failing in #1289:

1. test_min_max_quantization_graph[model_to_test3] (Maskrcnn-12)
   Solution: made sure that the updated reference .dot file has new quantizers in the expected places.
   Here's the proof:
   
   
   
   Add operation is ignored only.
   Matmul + Add is fused for ONNX: https://github.com/openvinotoolkit/nncf/blob/develop/nncf/experimental/onnx/hardware/fused_patterns.py#L56
   The entire fused pattern is not ignored, matmul is quantized, which is expected with the proposed changes Do not ignore the entire fused pattern if non-primary nodes are in the ignored scopes #1289.

2. quantization.test_algo_quantization.test_activation_ignored_scope[update_config_info1-should_ignore_quantizers1]
quantization.test_algo_quantization.test_activation_ignored_scope[update_config_info2-should_ignore_quantizers2]
   Solution: made sure that test produces quantizers in the expected places.
   ReLU is ignored in Conv + ReLU pattern, but Conv's input is quantized, as expected.