Currently, in Python, DynamicVamana.build always creates an index with float32 data, even when given an array with other types, such as float16 or int8. We should add support for the other NumPy types Vamana supports for building. Compressed non-NumPy types can be considered in future issues.