FlowBoundary is flow direction constraining

[visr]

(cherry picked from commit e84ae3a)

Fixes #1423

[visr]

Hmm CI doesn't like this. Perhaps this fixes #1423, but there are more tests to adopt it seems. @SouthEndMusic perhaps you want to take over this PR?

[SouthEndMusic]

Hmm CI doesn't like this. Perhaps this fixes #1423, but there are more tests to adopt it seems.

To solve the problem posed in #1423 you probably only need the change in utils.jl. The change in allocation_optim.jl can be kept here, that one makes sure that if the outflow edge of a UserDemand node is the reverse of its inflow edge, they are still always kept separate in the allocation flow output.

I did quite some digging to understand the change you took over in allocation_init.jl. I still don't really understand it. It is about how sources in the main network are handled, and for some reason I thought it was a good idea to treat sources in the main network as main-to-subnetwork connections. However, the consequences of this are quite complex and probably not as desired. I suspect that it comes from an observation that the capacity of sources in the main network is not taken into account properly, but this is not the way to solve it. We can make an issue for that now, or wait until we can inspect allocation flows in QGIS.