Text this: PreparedLLM: effective pre-pretraining framework for domain-specific large language models