[WIP] Fix `Model::extendConnectToModel` graph traversal

[adamkewley]

WIP

⚠️ This is a bigger diff than reality

This work is Related to #3694 . The diff here will be smaller once #3694 is merged. Otherwise, this PR will be refactored into an independent patch.

(I was too lazy to break this branch away from the feature branch, is what I'm trying to say).

---

This is to fix some of the bugs I've noticed in the multibody graph traversal part of OpenSim. Notably:

• It doesn't currently take the frame socket topology of Model into account, which leads to issues once you start doing things like chaining PhysicalOffsetFrames together or using StationDefinedFrames (which, themselves, depend on Stations, defined in PhysicalOffsetFrames, etc.)
• Doesn't account for other socket topologies in the model, which is why some components have to manually call setNextComponentInSystem (when the socket system of OpenSim should, in principle, be enough to figure this out)
• Segfaults under certain conditions, such as users creating unusual joint cycles in their models (e.g. https://github.com/ComputationalBiomechanicsLab/opensim-creator/issues/773). Those types of cycles are easy enough to create via a UI. They shouldn't result in a segfault (at worst, an exception is fine).

Brief summary of changes

• Refactor/tidy up Model::extendConnectToModel (so I/we can see what it's trying to do)
• Remove setNextSubcomponentInSystem calls that are peppered throughout Model::extendConnectToModel
• Centralize almost all setNextSubcomponentInSystem calls to a single for (auto& component : TopologicallySortedComponents(model)) {} expression

Testing I've completed

• Kicking it until it stops screaming, because it's a very central algorithm that's unlikely to squeeze past the test suite until every edge case is covered

Looking for feedback on...

CHANGELOG.md (choose one)

• no need to update because...
• updated.

---

This change is 

[adamkewley]

Closing: the problem this fixes is mostly fixed by the #3711 patch