Move PostProcessing onto Route Steps

[MoKob]

This pull request tracks the progress of moving the Guidance Post-Processing options onto the Route Step.

It consists of the following task:
-[]Adjust turn collapsing in assemble_leg.hpp: This process assumes name-ids to be the only decision that decides whether a turn is kept or not. This needs to be adjusted to compress only no-turns. Some instructions are actually required to stay on a road.
Since this is only a heuristic for the route name and would require a lot of changes, keeping the name as only source of naming the route seems like the better option.

• Move Post Processing away from PathData and past assemble_steps.cpp to work on route-steps. This would require the insertion of temporary route steps for silent instructions that can then be collapsed, or a full integration of the post-processing with assemble_steps. I'd prefer the former (@TheMarex, whats your take on this?)
• Post-Process Temporary Route Steps: Due to the adjustment above, some silent instructions might remain in the route-steps. Post-processing has to combine these accordingly.
  • we have to count exits for roundabouts
  • In addition, we can add information on intersections in between the instructions. Every silent turn can be added to a list of intersections with respective distances along the current step geometry (thank you for the proposal @TheMarex)
• clean up silent instructions -- remove silent instructions from output to prevent them from being shown in the json result.
  • remove silent instructions from output
  • potentially sync leg-geometry with new instructions?
• static turn instructions #2080 needs to be merged prior to this
• find out why path length and segments do not match
• get @daniel-j-h 's blessing

[MoKob]

@TheMarex I have a question regarding the leg_geometry.

After https://github.com/Project-OSRM/osrm-backend/blob/fix/guidance/postprocessing/src/engine/guidance/post_processing.cpp#L232-L241, which removes temporary instructions in post-processing to count roundabouts/intersections, the leg-geometry segments are out of sync with the actual turn-instructions.

Should we undertake the effort to set the correct indices again? Essentially translating the turn-instruction geometry indices back into the segments of the geometry? The coordinates itself should be unaffected.

[MoKob]

An additional item to discuss: to output the intersections in between, should we output the duration/distance along the currently looked at segment, or the distances/durations between the different intersections themselves?

So for (S) --a -- (I) -- b -- (I2) -- c -- (T), should we rather output:

Option 1: Depart S, (a+b+c), Intersections: [(I,a), (I2,a+b)]

or

Option 2: Depart S, (a+b+c), Intersections: [(I,a), (I2,b)]

In both cases, we would end up adding some additional information down the line on the intersection shape.

Edit: currently option 2 is implemented, both come with coordinate of the intersection

[daniel-j-h]

From what I can see I think you only need the instruction to pass it on (and you do not move it), so a const-ref (or const value since it should be cheap to copy it) should be fine, too.

[MoKob]

Might be related: #1575

[MoKob]

Path length in front-end are rounded to large differences (426->450). This results in the differences in the path data length I've seen. The values seem correct, though.

[MoKob]

@TheMarex this one is all yours

[TheMarex]

@MoKob looks great! Going to rebase and merge.