Solla: Towards a Speech-Oriented LLM That Hears Acoustic Context