[Docs] Add section explaining the architectural decisions of using flat trees instead of nested ones for structuring datasets

[farrajota]

Add a subsection discussing why all fields of annotation's data are converted into a flat tree instead of being nested like in their original dataset format and explain the benefits of using such strategy for either data visualization and dataset normalization for all kinds of tasks.