Improve default B-spline endpoint speed

Right now it's set to 2.5mm per t-unit. This is totally arbitrary and catastrophically bad in some cases. Either we should make `auto_speed` the default, or, if that's too expensive, do our best to estimate from the endpoints (and maybe waypoints and end tangents). This would obviously be breaking.

Start by locating the B-spline endpoint-speed default and the existing `auto_speed` implementation, then inspect how endpoint, waypoint, and end-tangent data are available. Compare the cost and behavior of the proposed alternatives. Done means the default behavior is chosen and implemented, breaking behavior is documented, and representative speed cases are covered by tests.

Written by the indexing model from the issue text.