Leveraging LLMs for Synthesizing Training Data Across Many Languages in Multilingual Dense Retrieval