Reset the center line to untrimmed before calling intersection_polygo…

[dabreegster]

…n. Will revisit this at some point. #136, closes #143.

[dabreegster]

The test diffs look fine.

Longer-term, we kind of have two approaches to decide between. Both should be compatible with the goal of always having a trimmed center_line and intersection.polygon filled out and not just calculating it last-minute in the transformation.

1. Keep assuming the input to the intersection geometry routines is untrimmed. I think we'll always have enough state around to manage that. The only thing to maybe figure out is that we never lengthen a center line that was shorter in a previous pass. That was partly the motivation for storing trim_start and trim_end explicitly.

2. Make the routines handle input that's already trimmed, and consider trimming further if something has changed. This feels harder to reason about to me, personally.