bug: min_signal=0 fails pgo

[hopeyen]

The default for signal is usually 100, which filters out subgraphs with zero signals. When an indexer wants to pinned subgraphs with 0 signals, they would lower the min_signal requirement and ran into errors pasted in the gist .
The most straight-forward solution may be excluding those subgraphs from participating in the optimization problem. Alternatively, (my intuition for modifying the algo) we would need to add additional handling for finding the zeros in the dual problem.