Text this: Enhancing Large Language Models for Underrepresented Varieties: Pretraining Strategies in the Galician-Portuguese Diasystem