[planning] Revise CollisionChecker parallel edge checks

[calderpg-tri]

From discussion with @tomstewart-woven and @aochiai, CollisionChecker's parallel edge checks can be improved by using the range-based parallel-for loops rather than the naive index-based versions. Using the range-based loops means less work is performed once a colliding configuration is found, since it allows the entire rest of a thread's range to be skipped.

Also contains two other revisions:

• MeasureEdgeCollisionFreeParallel now uses an implementation of the behavior of std::atomic<T>::fetch_min instead of separate std:atomic<double> + std::mutex.
• CheckEdgeCollisionFreeParallel now special-cases the q1 check so that it is performed before parallel dispatch.

+@jwnimmer-tri for review, thanks!

---

This change is 

[jwnimmer-tri]

I don't have a lot of free time these days. Maybe the users who are interested in this feature would like to feature-review it?

[sherm1]

This PR still needs a feature reviewer assigned.

[tomstewart-woven]

@jwnimmer-tri I'm happy to feature-review this if that's alright with you

[jwnimmer-tri]

Thank you!

[calderpg-tri]

Thanks!