Refactor flow integration

[SouthEndMusic]

The flow integration is now refactored so that integrated flows are computed based on water_balance! calls after the timestep but before parameter updates, so that the flows are calculated with the values of the discrete parameters that they had over the last timestep.

I also made sure that the db is closed properly in the tests.

This PR was originally intended to fix #1413, but that was put on hold because the resulting water balance errors were quite large, which will possibly be fixed by #1808.

[SouthEndMusic]

TL;DR for @HendrikKok : there are probably still problems with flow integration in the current Ribasim main.

In the interval of a single timestep, discrete parameters (from DiscreteControl, basin forcings) are fixed. That means that for accurate flow integration over that interval, all flows must be formulated with these correct parameters set. This is particularly important on the boundary point with the previous timestep, which we currently do wrong on main. To solve this systematically, flow integration should only happen with flows formulated in the single integrate_flows! call, which happens after a timestep and before the updates of discrete parameters for the next timestep.

I already know this isn't the holy grail for fixing the water balance error computations, because I experimented with integration schemes that satisfy the above and the large water balance errors are still there.

[SouthEndMusic]

This has been made redundant by #1819.