refactor: Graph & IterableGraph TOML roundtrip

[RobBuchananCompPhys]

There's a couple of things going on in this PR, all with the purpose of fixing deserialisation of IterableGraph from TOML:

• The problem was in the first instance that deserialisation builds out the graph in the 'opposite' direction to how we typically construct them manually*, such that when it comes to parsing a 'sub'-graph (such as any graph containing an instance of IterableGraph), the edges that depend on external nodes connected to the sub-InputsNode basically fail to materialise, since they have no awareness of any such data source. Upon closer analysis it was found that this was a problem occuring for deserialisation of any Graph node, iterable or otherwise. This issue has been fixed in this PR, so a TOML roundtrip is possible for the SubGraph test. This exact problem would have been the root cause of the IterableGraph loop edges not being parsed, however...
• ...it turns out that no specialised de/serialisation code had ever been written for IterableGraph and LoopEdge classes, so the loopEdges_ were never parsed anyway. This has also been fixed, and there are new tests for this.

*it should be noted that after implementing these changes, it should in theory be possible for a user to manually construct a nested graph from the inside-out without having yet emplaced the relevant external nodes! It may however require some further GUI work where, for instance, hovering an edge over a graph's InputsNode triggers the edge creation, whether or not any (proxy) inputs exist or not.

[rprospero]

This is going to conflict with my #2560, but I can do a rebase.

[trisyoungs]

One heavily suggested change, otherwise looks OK.