Parallelize the incremental indexing of large descendants

[irevoire]

In #116 a big part of the indexing process cannot be parallelized anymore.
We need to find a way to parallelize the function that explodes the large descendants into trees.

Steps to achieve multi-threading requires implementing the following issues first:

• Stop pointing to item directly in split nodes #119

[irevoire]

After meeting with many, we had multiple ideas:

• We can parallelize a single descendant by generating its tree (single thread) and then inserting all the remaining elements that don't fit in RAM in parallel. As long as the threads don't modify the tree in the database, this can be done in parallel
  => Cons: the first non-parallelized step can still take a lot of time, especially when we have a new tree to generate
• We can parallelize over multiple small descendants, as long as they fit in RAM, a thread can take descendants by itself.
  => Cons: When creating new trees, we're entirely monothreaded
• We can parallelize the common elements of multiple large descendants at the same time, but then it's not super clear how to do the subsequent insertions of different elements that don't fit in RAM

Whatever we move to, we'll need to implement #119

[irevoire]

Fixed by #130