Please see discussion in zulipchat.

This PR covers two important points:

1. The effectedSensitivity calculation needs to sum over the segments - otherwise a change in ISF can result in discarding previous segment results and thus the minimum correction calculation may be too high.
2. Calculate and apply a maxNonSuspendUnits - this is important for safety, for example in edge cases where there is a temporary higher target in play, which can result in some predicted values then having negative correction units (i.e., the weighted target is above the prediction). As a result, the final calculated dosage may result in glucose going below the suspend threshold in such regions. While not observed, it is definitely safer to implement a simple safeguard for this.