Remaining documentation improvements

[Samuel-amap]

Some leftovers not done during the documentation overhaul completed in #128:

• Some of the API inconsistencies could probably be detailed a little more (see API inconsistencies #131)
• Initialisation order needs more clarification. Status templates and variables are initialised / queried in a specific order : (If I'm not mistaken) soft dependency nodes initialise their scale's status templates. After that, nodes already present in the MTG are queried for any variables that may be initialised. If a status template is empty but some variables are requested as global outputs the corresponding template will be initialised [Note : this can leave some status templates uninitialised, see Uninitialised status template in HardDependency #103]. Then user-provided variables are inspected (?)
• Double-check if it is clear that there is no current way of handling cyclic dependencies in modellist mode :
• add_organ! and node creation with attributes need to be detailed a little, possibly
• Might be worth discussing further the fact that deleting nodes isn't recommended atm

Some fixups to additions made in #128:

• Toy plant : Properly check root growth is limited per timestep : it is the case, but a proper state machine would be clearer.
• Examples with multiple modellists
• Under the hood API tinkering example and dedicated documentation page -> get_nsteps, replace mapping status vectors, GraphSimulation (only partially done)
• Graph images illustrating PSE's graph specificities :(
• Examples : check Toy simulation water_absorbed output
• Add toy plant examples to the toy model testsuite ?
• Examples using the function to display a dependency graph
• Make the 'errors and troubleshooting' examples doctests
• Reintroduce an example to get the weather (was part of the model translation page, which has been removed in this PR)

[Samuel-amap]

Another idea that might be interesting for newer users :

A tutorial/physiologically plausible example with a specific process left as a blank slate, to serve as a hook for users interested to write a quick model to grind their teeth on PSE, and to showcase a realistic scenario where implementing/trying out multiple models can quickly enable testing a hypothesis.

[Samuel-amap]

The sections discussing PreviousTimestep should be reworked, and the structure renamed to OpenCycle, probably. The fact that it might cause extra instability (within the timestep, and not much more than other variables) in ordering for the variable when changing models should be mentioned as well.