feat: fmi3; nominals_continuous_states before initialization yields X…

[PeterMeisrimelModelon]

…ML nominals

Mirroring the principle behavior we already had in FMI2.

[efredriksson-modelon]

👍

Should probably have this defined as part of some interface in the future. It is very implicit that correct tolerances depends on a private property that is set calling a function triggering evaluation of this. For me it is not clear why the algorithm driver does not just call the FMU and ask for nominal values?

[PeterMeisrimelModelon]

👍

Should probably have this defined as part of some interface in the future. It is very implicit that correct tolerances depends on a private property that is set calling a function triggering evaluation of this. For me it is not clear why the algorithm driver does not just call the FMU and ask for nominal values?

Context here what happens in the algorithm driver on why one needs this:

1. If no absolute tolerance was specified, it is automatically set as 0.01rtolnominals. After initialization, so this will also yield the correct nominals.
2. Consider that atol was explicitly set to e.g., atol = factor*nominals, with factor != 0.01. This requires evaluating nominals before simulation/initialization. We check if nominals were evaluated before initialization via the _preinit_nominal_continuous_states attribute. If yes, we try to determine factor and adjust atol to atol = factor*<nominals_after initialization>

Not super pretty, but does the job. Can re-visit clearing this up a bit.