Run allocation first after BMI stop

[visr]

This allows allocation over period (t, t + dt) to use variables set over BMI at time t.

We no longer run allocation as a callback, but call it ourselves, such that we can control that BMI runs allocation first before running the physical layer.

[visr]

Some pretty large allocation test differences suggest we don't call allocation correctly yet.

No Captured Logs for test item "Allocation level control" at test\allocation_test.jl:308 on worker 26268
Error in testset "Allocation level control" on worker 26268:
Test Failed at D:\Ribasim\core\test\allocation_test.jl:334
  Expression: ≈(storage[stage_1], u_stage_1.(t[stage_1]), rtol = 0.0001)
   Evaluated: [500.0, 672.7963463605041] ≈ [500.0, 543.2] (rtol=0.0001)

[SouthEndMusic]

This might have to do with the fact that allocation uses the flows from the last water_balance! call, which depends (maybe among other things) on the used solver and might not correspond to the right state of the Model. To make this consistent, maybe always do a water_balance! call before allocation, which also would help with performing allocation at t=0 (#1389).
Alternatively, we should first pick up the issue about using average flows and levels as input for allocation (#1121, but this issue does not yet take basin levels into account).

[visr]

It actually ran allocation on the wrong times, that is fixed now.

[SouthEndMusic]

Looks good, left some small comments

[SouthEndMusic]

Now that you have this, can it be removed below?

Ribasim/core/src/model.jl

Lines 50 to 53 in e10f657

	t_end = seconds_since(config.endtime, config.starttime)
	if t_end <= 0
	error("Model starttime is not before endtime.")
	end

[visr]

Right, I forgot, thanks.

[SouthEndMusic]

I'd say the more common way to do this is

if t % config.allocation.timestep ≈ 0.0
   ... 
end

and the other check could just be t > 0 if you actually need that.

[SouthEndMusic]

Although the comparison tolerances work out differently

[visr]

That would only work if the t is right after, but not right before, right?

julia> prevfloat(3.0) % 0.5
0.49999999999999956

[SouthEndMusic]

Oops

lol