[Reconversion] Get back to rangeToStructure()

[Reinmar]

Part of #10294.

• Recap of what we had.
• List of problematic cases. Write them down.
• Reflect on how it applies to the new reality.

[niegowski]

The old PoC was in i/9407-list-reconversion branch with:

• A first rough implementation of re-conversion with the usage of slots. It also included slotFor( 'self' ) to be able to trigger the conversion of the attribute that results as a wrapper element around the element with the converted attribute. This was mainly rewritten in the feature branch. The slotFor( 'self' ) is not yet ported.
• MappedRange (temporary name) - The concept of a collection of flat ranges that update their start and end according to the Differ output. Those are reacting to a specific list of attributes to keep a consecutive list of list elements as a single list. This part is working pretty well, it also has automatic tests. This is supposed to track where are the ranges considered as a virtual items in the model. Those were expected to reduce changes provided by the Differ to replace some of them with the re-conversion items (but failed in cases of joining/splitting lists).
• The Differ providing more details for the DiffItem: element reference and map of attributes. This was used by MappedRange to find out which changes should affect the range.
• Some prototypes of low-level conversion handlers for the lists (a base for future downcast helper).
• The removing of view elements was a hybrid solution that used element/list reference to remove the whole view element (instead of mapping positions before/after the removed element).

The biggest problem (besides the roughness of that implementation) was how to reduce Differ's changes so they would contain insertRange and removeRange in the proper sequence.