Optimize Finder.get_siblings/2 memory usage and speed

[preciz]

Replaced a multi-pass pipeline (Enum.reverse |> Enum.drop_while |> tl |> Enum.filter) with a single-pass tail-recursive approach.

Since children_nodes_ids are stored in reverse document order, iterating over them and prepending to an accumulator naturally reverses the relevant subsequent siblings back into forward document order without needing to iterate the preceding siblings or create intermediate lists.

Benchmark results:

Name                       ips        average  deviation         median         99th %
new_get_siblings         89.90       11.12 ms    ±14.39%       10.97 ms       15.05 ms
old_get_siblings         41.31       24.21 ms    ±10.01%       24.15 ms       31.30 ms

Comparison:
new_get_siblings         89.90
old_get_siblings         41.31 - 2.18x slower +13.09 ms

Memory usage statistics:

Name                Memory usage
new_get_siblings         2.97 MB
old_get_siblings        10.31 MB - 3.47x memory usage +7.34 MB