Collision calculation has a bug

[tskisner]

As part of my work re-factoring / re-writing fiberassign, I discovered a bug in the collision calculation. The central body and ferrule holders are modeled as circles and segments, and overlap of these shapes are used to detect collisions. For mult-point segments, each subsegment must be checked for a collision. However, the current code returns immediately after checking only the first piece of a multi-point segment:

https://github.com/desihub/fiberassign/blob/master/src/collision.cpp#L186

For verification of the problem, using the small test data set, tile 1148, fiber 1597 is assigned to a target which produces the configuration in the attached image. My new code identified this as a collision, but the master version did not. I'll open a PR in a moment to fix this in master, even though the code will be replaced shortly. The consequences are that any previous collision calculations involving the multi-point central body segment and the ferrule holder head, where the head crosses segments other than the first one, would be identified as not colliding.