This makes the code a bit more consistent and lets us avoid some
hacky and probably non-performant things:
- abusing CSS classes in the draw/drag datum functions (classed `.target`)
(is this thing target? just check d.properties)
- regexing the id for `-nope$`
(is this thing a nope target? just check d.properties)
- using context.hasEntity to get a the real entity
(is this thing a real osmEntity? just check d.properties)
- fixes code like the restriction editor which uses fake ids for split ways
Drag start is responsible for switching into drag mode, classing
stuff as `active` and kicking off a bunch of other things.
If the drag move happens immediately after this, and includes the
target from the initial active drag, it can cause weird snapping
from the dragnode to its own parent way. (Happened if the user did
a very fast drag from the node along the parent way just next to it)
This is because #surface can be transformed during a partial redraw
and d3.mouse() returns transformed point, not the real screen point.
a2cf4f32ec/src/point.js (L4-L9)
