Previously, we only support float64. We now support float32 and float64. When
constructing a tensor without providing a data type, the default is float32.
Fix the tests to data type consistency. All PyTACO application tests now use
float32 to match the default data type of TACO. Other tests may use float32 or
float64.