Fix the post-processing of chain-rule sampling on specific wires for Gaussian backend (#120)

[Hugh-888]

• Description:
  Bug for chain-rule sampling on specific wires Chain-rule sampling is incompatible with measurement on specifc wires #120.

• Root Cause:
  The implementation reused the same post-processing method as in the MCMC case. However, the sample structures of the two methods are different.

• Solution:
  Adding an additional post-process method for the chain-rule case.

[Hugh-888]

• Test correct Fock probability when setting with_prob=True with chain-rule method.

nmode=3
cutoff=3
data = torch.rand(2,2)

cir = dq.QumodeCircuit(nmode, cutoff=cutoff, init_state='vac', backend='gaussian')
cir.s(wires=[0], r=1, theta=0)
cir.s(wires=[1], r=1, theta=0)
cir.bs(wires=[0,1], inputs=data[0])
cir.bs(wires=[1,2], inputs=data[1])
cir()
samples = cir.measure(wires=[0, 2], shots=30, mcmc=False, with_prob=True)
samples

cutoff2 = 11
cir2= dq.QumodeCircuit(nmode, cutoff=cutoff2, init_state='vac', backend='gaussian')
cir2.s(wires=[0], r=1, theta=0)
cir2.s(wires=[1], r=1, theta=0)
cir2.bs(wires=[0,1], inputs=data[0])
cir2.bs(wires=[1,2], inputs=data[1])
prob_dic = cir2(is_prob=True)

for key in samples:
    prob = 0
    for k in range(cutoff2):
        temp = torch.tensor([0]*3)
        temp[1] = k
        temp[0] = key.state[0]
        temp[2] = key.state[1]
        s = dq.FockState(temp)
        if s in prob_dic:
            prob = prob + prob_dic[s]
    print(key, 'result from partial cov:', samples[key][1], 'result from full cov:', prob)

[sansiro77]

LGTM