accessibility: Validate and document the behavior for accessibility at arrival

[tahini]

After adding some unit tests to the accessibility map for the backward calculation (with arrival time), it seems the travel time is calculated between the requested arrival time and the departure time of the transit from that node, excluding the waiting time.

1- Is that correct?
2- Should it rather be the arrival time of the transit - waiting time?
3- How should those results be interpreted from the consumer's point of view? For its calculations, Transition seems to take the wanted duration of the trip minus the travel time to calculate the remaining time and get the accessibility radius. But if it excludes waiting time, doesn't it over-calculate the accessibility?

cc @kaligrafy

[tahini]

So the times at node should include the waiting time (transfers include waiting time, calculation from a departure point include the waiting time). So it would be the transit departure time minus waiting time.

But the accessibility calculation endpoint should support a waiting time of 0 if we are interested in perfect transit time (that may already be the case).