Fix allocation flow output

[SouthEndMusic]

What
It seems that there are some problems with the allocation flow output in allocation_flow.arrow:

• There is no guarantee that the edge the data is written to is an existing edge in the input (from, to) instead of its non-existent reverse (to, from);
• Data for bidirectional allocation edges should be combined into a single output row
• Use average flows over allocation intervals for realized. In the case of level demand, use the difference in storage between the beginning and the end of the basin. Moved to new issue: Use average flows over allocation intervals for realized in allocation.arrow #1356

How
Per task above in save_allocation_flows!:

• The existence of a flow edge (from, to) can be checked by its presence as a key in graph[].flow_dict (or is such a lookup unnecessarily slow?);
• Use that when both directions of an edge exist, they appear consecutively in graph[].edge_ids[allocation_network_id] and thus also in the index set of problem[:F] which is looped over in save_allocation_flows!.
• For the integrated flows: integrate flows in integrate_flows! for the flows that are relevant for this allocation input, stored somewhere in the allocation data structure (make sure that this goes well in subnetworks where all optimizations are run twice.). For the storage difference: use the current u and the previous one which can be taken from the integrator object (which is available in the update_allocation! function).

[SouthEndMusic]

This should be fixed before #1066

[SouthEndMusic]

@SnippenE this is not the issue about the allocation output discussed at the last samenwerkingsdag. That one is already fixed: #1347. There are 2 different allocation output files:

https://deltares.github.io/Ribasim/core/usage.html#allocation---allocation.arrow
https://deltares.github.io/Ribasim/core/usage.html#allocation-flow---allocation_flow.arrow