Finetune step_limiter!

[SouthEndMusic]

Follow-up of #1911.

Fixes #1897 (?).

Fixes #1838 (how could we test that negative flows where they should not be no longer occur? Maybe add a callback that explicitly checks for non-decreasing states)

I focus in this PR on UserDemand inflow and infiltration because of their importance in coupling.

[SouthEndMusic]

@evetion you might appreciate this first commit; it gets rid of many of the magic hardcoded numbers for threshold values.

[SouthEndMusic]

@rleander this is an interesting one for you, this one together with #1911 should fix the problems you noticed, in particular: infiltration and realized user demand should no longer drift away from their demands as the simulation goes on (when the basin does not run dry).

[evetion]

Some general comments:

For the next time, please refactor in separate PR's, mixing it makes the PR and the git history in the future hard to read. I do like the refactor, tackling the magic numbers and the basin complexity 👍🏻.

Could you provide a comment in the PR description on what you introduce and why? I'll give more detailed comments in the code below.

After reading it completely, my main confusion is that we clamp the minimum flow out of a basin and not the maximum flow? I thought the reduction factors trigger so a (maximum) flow cannot empty a basin.