Add support for HOP4 attention block during conversion

Add support for attention block with reshape depending on shape.

• Several unit tests are added to robustify shape, reshape nodes cases. Correct matrix 4d unit test
• A special pattern detection is added for ONNX producing a sub graph which can be reduced to a single shape node.
• clean/refactor the code to support more transpose cases
• add support to output graph where inputs are named inputs (default values is put in place)
• reshape node conversion extended where shape is coming from a tensor
• transpose node is extended to detect user explicit transpose in case nchw of dim==4