[ MO Interpolate ] Fixing broken model reshape-ability

[jane-intel]

Description: Interpolate operation with hard-coded (constant) target_spatial_shape input usually prevents user from successful spatial reshape of the model. It is totally valid, but customer may not be able to change the original model, so we introduce two transformations for making Interpolate operation reshape-able:

1. InterpolateConcat -- is valid for all user scenarios and is enabled in MO by default
2. InterpolateReshapeWA -- is not valid in all possible cases, but as we observed on practice -- this transformation helps in 99.99% cases.

Ticket: CVS-32096 -- there may be more

Code:

• Comments
• Code style (PEP8)
• Transformation generates reshape-able IR
• Transformation preserves node names: N/A -- we replace constants. preserving their names is sort of weird as we probably already lost its original name and in case of reshape this nodes output will be different

Validation:

• Unit tests: Units are too small to test
• Framework layer tests: N/A
• Transformation tests: Done
• e2e model test with an update of reshape_utils.py: N/A
• MO IR Reader check: N/A

Documentation:

• Supported frameworks operations list: N/A
• Supported public models list: N/A
• New operations specification N/A
• Guide on how to convert the public model: N/A
• User guide update: N/A

Other:

• Sample/Demo application to infer the model: N/A

[slyalin]

Leaving code review for @lazarevevgeny, I approve the concept of "smart reshape". We should think whether we need to call transformations of this kind in run-time transformation pipeline in nGraph and where the best palace to do that (it should be done before network reshape). @jane-intel, please formulate your thoughts about it in that ticket.

[rkazants]

review finished