Fix for Reduce extractors and normalizer

[lazarevevgeny]

Description: Fixed extractor for Reduce operations from ONNX. If the axis value in the Reduce op is not specified then the extractor creates a numpy array the following way: int64_array(None). This value is not None because it is a numpy array. So the ReduceAxisNormalizer creates a Const with int64_array(None) value. This causes an issue in the transformations where the value of Const is compared with some other floating point values because numpy tries to compare None value and some floating point value.

JIRA: 43166

Code:

• Comments
• Code style (PEP8)
• Transformation generates reshape-able IR
• Transformation preserves original framework node names

Validation:

• Unit tests
• Framework operation tests: no new op enabled
• Transformation tests
• e2e model test with an update of ./tests/e2e_oss/utils/reshape_utils.py: N/A no new model enabled
• Model Optimizer IR Reader check: N/A since the change does not affect MO IR Reader

Documentation:

• Supported frameworks operations list: N/A no new op enabled
• Supported public models list: N/A no new model enabled
• New operations specification: N/A no new op enabled
• Guide on how to convert the public model: N/A no new model enabled
• User guide update: N/A no new user-visible change