Do not initialize torch by default and change set_precision arguments

[FHof]

• Support float32 and float64 in set_precision
• Replace "double" and "float" with "float64" and "float32" in the tests
• Do not import torch by default and add the set_up_backend function
• Remove the unused device argument from enable_cuda and change its data_type argument