Phantom node weights not being joined when collapsing match legs

[danpat]

It looks like when we're collapsing match legs together (triggered by use of the waypoints=0;N parameter), we're discarding properly adding the forward/reverse_weight part of the PhantomNodes that are not waypoints.

    @match @testbot
    Scenario: Regression test - duration aggregation for phantom nodes
        Given the profile "testbot"
        Given a grid size of 10 meters
        Given the node map
          """
          a--1-b2-c-3--d
          """
        And the ways
          | nodes |
          | ab    |
          | bc    |
          | cd    |
        Given the query options
          | geometries | geojson |
          | overview    | full |
          | steps       | true |
          | waypoints   | 0;2 |
          | annotations | duration |
          | generate_hints | false |
        When I match I should get
          | trace  | geometry                                   | a:duration | duration |
          #| 13     | 1.000135,1,1.000225,1,1.00036,1,1.00045,1 | 1:1.5:1    |          |
          | 123   | 1.000135,1,1.000225,1,1.00036,1,1.00045,1  | 1:1.5:1    | 3.5      |

The reported result for this test is currently:

     | trace | geometry                                  |     a:duration |     duration |
     | 123   | 1.000135,1,1.000225,1,1.00036,1,1.00045,1 | (-) 1:1.5:1    | (-) 3.5      |
     | 123   | 1.000135,1,1.000225,1,1.00036,1,1.00045,1 | (+) 1:1:1      | (+) 3        |

Note that the middle duration annotation is off by 0.5 - this is the amount of the forward_weight of the target_phantom of the first leg. The total leg duration (3) is also incorrect, as it is also missing the lost phantom weight value.

[danpat]

Fixed by #4949