Support multi-layering in FlowNet

To support the goal of training a FlowNet model on the Norne Open Dataset, horizontal layering is identified as a potential enabler.

To following steps have been identified:

[ ] (Issue: #240) Create function that can either create a 3D or a 2D field of well connections (collapse all connections to 2D) + tests
[x] (Issue: #235, PR: #246) Update Mitchels Best Candiate Algorithm to 2D cases + tests
[ ] (Issue #242) Create function that allows for 2D Delaunay triangulation + tests
[ ] (Issue #240) Create function to copy and shift 2D triangulation to desired k-layers from original model + tests (mimicking "seismic")
[ ] (Issue #243) Create a function that
- identifies "near vertical" connections based on a max angle --> fault throw (or use existing fault detection?).
- Remove connections across faults with a large fault throw.
- Find "horizontal connections" across faults and assign a fault multiplier.
[ ] (Issue: #239) Tie-up everything and allow the user to specify which k-layers should be used as FlowNet layers

equinor / flownet