Implemented smart hashing of batch updates for LMT

[rpanic]

This PR introduces a new batch-setLeaf algorithm that optimizes the amount of hashes we have to do.
The tldr on this is that if we set each leaf after the other, we do $height$ amount of hashes, but we can be smart about it and go level by level, in the process merging paths of leaves where they converge.
This is less useful for the previous tree implementation, where the leafs are randomly distributed and therefore very sparse. But for the LMT, where we place leafs one-by-one strictly incrementing their tree index, big parts of the paths will overlay, therefore giving us a decent speedup for batch updates.

LMT with batch update vs. vanilla LMT
100 transfers with unqiue state slots:

result generation:
5357ms vs 6987ms
therefore roughly a 30% speedup