allow user to define result dtype of dark / flat reduction

float32 should be considered instead of float64 for saving reduced images, especially if the files are not links. The safest would be something like reduce(..., dtype=numpy.float32) in the API.