Fixed scalar creating in tf.Graph decoder

[popovaan]

Root cause analysis:
In tf.Graph decoder scalars are wrapped to list, then to nd.array before passing to ov.Tensor() constructor.
Creating of intermediate nd.array leads to creating of additional object in memory which is deallocated before C++ part access it.
This leads to corruption of scalar value.

Solution:
Creating of additional list is redundant.
Intermediate nd.array needs to be saved to TFGraphNodeDecoder field to avoid deallocating.

Ticket: 117250, 119142
Closes #19019

Code:

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

Validation:

• Unit tests
• Framework operation tests
• Transformation tests
• Model Optimizer IR Reader check

Documentation:

• Supported frameworks operations list
• Guide on how to convert the public model
• User guide update