Handle multiple versions of the same leg better

[Olly Betts]

Pockettopo measures 3 versions of each leg, and if you feed such data into cavern it treats each version as a separate leg, producing a very loopy network which takes a while to solve.

Ideally we want to average the repeat readings in this case, and then solve the network treating each set of such readings as one leg. That would also make the leg count saner.

Not sure if we just always do this when we have multiple readings between two points, or if we only consider consecutive readings. The later is probably easier to implement if nothing else.

[Olly Betts]

If you repeat a leg consecutively, 1.2.17 averages the readings, adjusts the variances appropriately, and treats them as a single leg.