Incorrect response to stochastic point source

@muratore and @ohartwig pointed that the response to a stochastic point source was incorrect, as we expect something proportional to f and we get a flat response (see below).

After investigations, @ohartwig noticed that we call compute_hplus() and compute_hcross() multiple times in compute_gw_response(), and that the implementation of these two function is such that we draw new samples every time -- loosing all correlation. Their implementations indeed read

def compute_hplus(self, t):
    logging.info("Compute +-polarized gravitational-wave strain in the BCRS")
    return numpy.random.normal(scale=self.stddev, size=len(t))

def compute_hcross(self, t):
    logging.info("Compute x-polarized gravitational-wave strain in the BCRS")
    return numpy.random.normal(scale=self.stddev, size=len(t))

This should be fixed ASAP.