Context-Aware Dynamic Chunking for Streaming Tibetan Speech Recognition