Revisit gate fusion for simulation performance improvements

[stavros11]

The latest Qiskit state vector simulator shows particularly good performance for various circuits when the fusion_enable flag is True. We should revisit gate fusion for Qibo and check if we can improve performance. If this is observed in multiple circuits we should consider enabling the fusion by default for circuits larger than a pre-defined qubit number (for Qiskit this is 14 qubits).

We could also consider disabling multi-threading CPU (by default) for circuits smaller than 14 qubits, if the single thread performance is better in these cases.