fix mincut (fixes #324)

[etiennedeg]

When a vertex is merged during the cut of the phase, if the current vertex is not kept, we need to switch to the kept vertex

fixes #324

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (5878e7b) 97.30% compared to head (9b26193) 97.30%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #327   +/-   ##
=======================================
  Coverage   97.30%   97.30%           
=======================================
  Files         115      115           
  Lines        6796     6796           
=======================================
  Hits         6613     6613           
  Misses        183      183           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gdalle]

@etiennedeg did you check out #325? haven't had time to review yet

[etiennedeg]

Yes, his implementation drops the heuristics but the heuristic was not at fault. I noticed that he struggled to understand my code, so I added more comments

[axsk]

It's not true I did not drop the heuristic.

What is true that I did not understand is the whole algorithm, esp. the part with the cutweight -= ew (still cryptic to me) which is why removed those parts to bring it more in line with the original Stoer-Wagner Algorithm (which I do understand) but WITH the heuristic still applied.

Looking at this fix I could imagine it being still necessary for my patch though..

[axsk]

So I agree that updating u to the merged root should be done and have updated my PR accordingly.
I would appreciate if you could give it another look @etiennedeg

[axsk]

If you come to the conclusion you don't like my simplifications bringing it in line with Stoer-Wagner or they are not working with the heuristic let me know so I will close my PR and prepare a seperate one with the performance fixes alone.

[etiennedeg]

Closing in favor of #325 (see #325 (comment))