Leveraging fine-mapping and non-European training data to improve trans-ethnic polygenic risk scores