Possibility to generalize Backend.set_precision

[renatomello]

Right now, the Backend.set_precision allows one to choose between complex128 and complex64.

However, for some applications (e.g. ground-state estimations using VQEs) it can be beneficial to set the precision to float64 due to both states and circuits only containing real values.

In light of that, it'd be interesting to extend Backend.set_precision to more dtypes