[MO] enable ir_reader read and write Quantico model

[bszmelcz]

Overview:

• [x] enable MO ir_read read and write Quantico.onnx model

Detailed Changes

• add support for FP64 to ir_engine - resolves the error: KeyError: 'FP64' while using restore_graph_from_ir function
• update if statement to check if dim is not integer to prevent error: TypeError: argument of type 'int' is not iterable while restoring graph that was previously restored (restore_graph_from_ir) and saved (save_restored_graph)
• In If.py add unmask_shape to umask possible dynamic shapes since otherwise if the shape is dynamic the line graph_contain_fake_outputs = graph_contain_fake_outputs and np.any(out_node_shape == 0) produces output masked which later on produces the error TypeError: ufunc 'bitwise_xor' not supported for the input types in check if then_contains_fake_outputs ^ else_contains_fake_outputs.

Affected models/IRs:

• [x] Quantico.onnx => fixed
• [x] speech_transformer_encoder.onnx => fixed
• [x] if_diff_case => works
• [x] if_body_without_parameters => works
• [ ] Longformer => at the moment does not pass through MO so there is no way to check if ir_reader works for ir

Tickets:

• 82253

[rkazants]

@bszmelcz, could you please elaborate the problem in the description that you are solving?

[bszmelcz]

@bszmelcz, could you please elaborate the problem in the description that you are solving?

Sure, I am gonna update description and address the rest of the comments tomorrow morning