Text this: SkelFormer: An adaptive hierarchical transformer-based approach on skeleton graphs for human action recognition in video sequences