Optimize faust_torch module to avoid cloning data (by default) when possible.

It's still possible to clone data using the new argument 'clone' set to the value true.

Unfortunately it's mandatory to clone indices for a sparse matrix to sparse tensor conversion because torch imposes to use the same size types for both values and indices (even if values are floats and indices integers) but Faust/eigen uses 32 bits for indices and 64 bits for data (when double is used as FPP template parameter)

[skip ci]