Add conditional homodyne measurement

[Hugh-888]

Update details

1. Add homodyne gate circ.homodyne
   
2. Support homodyne measurements on arbitrary wires via arbitrary angles.
3. Access the sampling results
   
   if the circuits don't contain homodyne , circuit.measure_homodyne still works, but with the ideal
   multinormal distribution sampling.
4. Support the global gaussian state collapse when performing partial homodyne measurements.
   

[sansiro77]

可以参考我这个函数去简化一下forward函数。应该是不需要置换函数的，甚至连xxpp和xpxp之间的转换都不用。

from torch.distributions.multivariate_normal import MultivariateNormal

def measure_dyne(cov, mean, idx, eps):
    nmode = cov.shape[-1] // 2
    cov_m = torch.diag(torch.tensor([eps**2, 1.0 / eps**2] * (len(idx) // 2)))
    idx_rest = np.delete(np.arange(2 * nmode), idx)

    cov_a = cov[idx_rest][:, idx_rest]
    cov_b = cov[idx][:, idx]
    cov_ab = cov[idx_rest][:, idx]

    mean_a = mean[idx_rest]
    mean_b = mean[idx]
    mean_m = MultivariateNormal(mean_b, cov_b + cov_m).sample([1])[0]

    cov_out = cov_a - cov_ab @ torch.linalg.inv(cov_b + cov_m) @ cov_ab.mT
    mean_out = mean_a + cov_ab @ torch.linalg.inv(cov_b + cov_m) @ (mean_m - mean_b)

    return cov_out, mean_out, mean_m

[sansiro77]

目前这个Homodyne没有相关的指定shots的功能。我的想法是Homodyne的forward只返回[cov, mean]，用self.samples去记录采样结果。其实输入的cov如果是(batch*shots, 2n, 2n)，Homodyne就已经处理了shots不为1的情况。
QumodeCircuit里的self.state只记录operators的结果，self.measurements的输出赋值给临时变量或者记录为self.state_measured之类的，并且把计算过程放到measure_homodyne里，自动判断或者加个参数来判断执行ideal还是conditional Homodyne。
Generaldyne和Homodyne继承自Operation就可以了，放到新文件measurement.py里更清楚。

[sansiro77]

Good job!