Lane count transition regression

[Zero3]

I noticed a regression in lane count transition drawing since the previous version of map-machine on the python3.8 branch. It used to look pretty good:

But now it looks broken in many places:

It seems to happen when two ways are joined with a point, and the lane counts of the ways are different.

[enzet]

Hi, @Zero3!

Yeah, there was a number of compromises I made before I can do things the right way. I mean, it is a regression, but it is an intentional one. I removed connection circles to get rid of some other bizarre bugs. Hope I can fix it soon.