Simplify and optimize operators

The goal of this MR is to simplify and optimize operators through:

• Code factorization, with the introduction of the template OperatorTensorWithImpl;
• Strings elimination when not necessary, through the use of constexpr descriptors only;
  • Removed C++ getInputsName(), getOutputsName() and attributesName(), but their Python binding remains.
• Removed deprecated Scaling operator;
• Moved generic TransposeImpl in its own file like other generic operators.