> the software could be tuned so that after finishing the flood stage it should not try to turn the water off but look for the next stage. I say this because i see a large water flow drop after the flood stage
You should see that drop. Then it's working as intended.
In leva!'s terminology, flooding is the pressure-less stage before profiling can kick in. Its only purpose is to fill tubes and other air pockets with water, not brewing or preinfusion. The flow rate - of clear water, not espresso - in that stage is of no importance.
There can be no pressure profiling while there is easily compressible air in the fluid system. A vibe pump achieves little to nothing for a few seconds, pressure-wise, due to that air. The system must be filled almost entirely with water for pressure profiling to become possible. That's why the pump is run with full power until some pressure (0.7bar) appears.
At that point, pressure profiling becomes possible and the control loop kicks in. If you have a profile that begins at 0.7bar (or below, see "setpoint stitching" in manual), then the error in that moment is zero - because the pressure is where the control loop wants to have it. If the error is zero, the proportional output is 0. The integral output is 0, too, at 0s. Only the bias remains. As a logical result, the flow rate drops instantly and dramatically. If you see that drop, the control loop works as it must.
If you don't like the audible stutter generated by the transition from flooding to pressure control, start your profile at a higher pressure than 0.7bar (e.g. 1.2bar). In that case, the control loop can not let the output drop as much. You might want to try a different flood pressure setting, too.