bug: can't do search with *_FLAT over float16/float64 vectors

[BubbleCal]

now we hard code the type as float32. because we are trying to avoid dynamic dispatching for performance, need to figure out a way to get this work