Factor graph with all outputs plugged

[rafaelha]

For sampling, we compute marginals. The very last diagram is the computation P(x_1=s_1 x_2=s_2 ... x_n-1=s_n-1 x_n=0). This has all outputs plugged. Here the normal and conjugated halves (of the doubled diagram) are decoupled.

In the current approach, these two decoupled diagrams are still evaluated as a whole, effectively doubling the number of T gates.

Requirements:

• Evaluate only one half of the doubled diagram
• Figure out how to deal with normalization.
• Also allow construction of marginal probabilities from this diagram. This is because diagrams with few traces have less T gate cancellations. Consider having an automated threshold for this parameter.