Running shortest_graphlike_distance slow at large circuits/large cycles

[gabriellelcr]

Perhaps not a surprise, but stim_circuit.shortest_graphlike_error() is slow at large cycles/large distances (60 seconds for 1_000 rounds at distance 11 for a standard surface code). Is there a way to allow the function to recognize if there's one big loop/some repeating structure, and take less time?

We can get around the timing problem by just running shortest_graphlike_error() on the same circuit with a smaller number of cycles, but it's a convenient feature to be able to run it blindly on any circuit (eg when generating metadata for an experiment).