Add support for lower than 8-bits precision

See aidge_export_cpp!69 (merged).

Main change in this MR: adaptToBackend() improvements:

• Do not introduce a Transpose operator if not necessary;
• Ability to chain data type conversion and data format conversion;
• Changed adaptation cost to take into account the different type conversion.

This MR also includes several small fixes:

• Cast all types to double in main_compare.jinja, to allow comparison of integers and low-bits precision integers;
• Added sub-8-bits types in export data converter;
• Added isUnsigned(), isPacked() and getUnderlyingType() methods in DataType;
• getUnderlyingType() is now used to check the correct data type in tensors set and get;
• Add debug outputs in case of failed GraphView::replace();