Make signature of torch compilation target identical to func/numpy/sympy targets

[jakobj]

Since #294 there seems to be a difference between how the torch module is called and the other "compiled" callables are used. This should be unified by adjusting the generated torch module code.