Left-to-right expand observed equations into defaults during initialization

[hersle]

Fixes #2859.

This slightly changes how observed equations are expanded into a map of defaults when an ODE is initialized with defaults. Before this process took rhs => lhs by default, but flipped it to lhs => rhs if rhs was a parameter (my interpretation: "if rhs is something known").

Now it instead takes lhs => rhs by default, and flips if lhs is something known. I think this makes more sense, because unlike the arbitrarily complicated rhs, lhs of an observed equation (in a simplified system) is guaranteed (?) to be something simple (e.g. 0 ~ rhs or x ~ rhs). It is therefore better to decide whether to flip based on lhs.

I am not sure I have thought of everything. But it looks like all MTK tests are passing on my computer.

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

[hersle]

Also, should isparameter(lhs) always fail now? There cannot be a parameter on the left side of an observed equation?

[hersle]

Is the failure because of SciML/ModelingToolkitStandardLibrary.jl#306?

[hersle]

Also trying to remove the (thought-to-be) redundant isparameter(lhs) check now.

[ChrisRackauckas]

Also, should isparameter(lhs) always fail now? There cannot be a parameter on the left side of an observed equation?

Yes

[ChrisRackauckas]

The tests covered all of the reasons for flipping in the first place, so if this is passing then it's no longer required. This intuitively makes more sense and is simpler too, so it seems strictly better. I can't see a system where it would fail. I believe part of this was due to dummy derivative handling, but since we greatly improved that in other ways we seem good now.