After noticing some strange results in the $k$-nearest $p$-median implementation, I have realized that we had completely omitted the generation of opening constraints in the model. While the model will still solve (and potentially optimally) in certain scenarios, the results are erroneous:

• some clients may not be allocated to a facility
• the $p$ constraint not respected, since clients can be allocated to any facility

I have verified with diagnostics and plotting.

xref #414