Faster to_list: 20X for small Vectors and 8X for large ones

Custom implementation of to_list.
Recursively walk the trie in reverse order (to build the resulting list in the right order).
Adaptive strategy for prepending leaf blocks:
* for Vectors up to 3 levels deep (up to 32,768 elements) use Tuple.to_list and list concatenation
  (up to 1,024 temp lists created)
* for bigger Vectors use custom prepend one by one function as the overhead of creating and reverting
  lists becomes significant

More informative description.
Ability to run benchmark in `full quick` mode.