Ad-hoc graph introspection instead of Monitor nodes for thumbnails and click targets

Latest plans call for removing the *Monitor* node approach of extracting thumbnail data and instead executing the graph to obtain the thumbnail with specific bounds that it prefers, as well as potentially other `Context` values specific to the individual render of the thumbnail (or a user-requested thumbnail render from within the dialog that shows up when expanding a thumbnail into a graph value diagnostics interface).

---

Old issue description when this was titled "Automatic monitor node insertion":

In order to display thumbnails for every wire, a Monitor node should be placed at outputs during compilation. The number of monitor nodes should be minimized to avoid redundant clones as data flows through the graph. The outputs would be selected based on which thumbnails should be rendered, both in the node graph UI and layer panel. This would also improve #2028 and getting artboard translations in the [Clear Artboards PR](https://github.com/GraphiteEditor/Graphite/pull/2015).

Monitor nodes will be the standard way of getting data from the node graph, as opposed to the render output. This is because the render output does not include all data in the node graph due to destructive operations like rasterize or boolean operations. We currently work around this by redundantly storing the upstream graphic group for vector data created by a boolean operation, but this is not ideal. (This is something that is still currently being discussed)

https://discord.com/channels/731730685944922173/731738914812854303/1291860544604602439
https://discord.com/channels/731730685944922173/731738914812854303/1291830448195899527

No se nombran archivos ni pruebas. Empieza rastreando las rutas actuales de thumbnails y objetivos de clic de Monitor-node, y luego compáralas con la ejecución del grafo y los valores relevantes de Context; el trabajo estará terminado cuando el enfoque acordado sustituya la extracción ad-hoc de Monitor y cubra los thumbnails de la UI, los objetivos de clic y los renders de diagnóstico solicitados.

Escrito por el modelo de indexación a partir del texto del issue.