Just think of each node as an extensible component, which can have more children extensible components. A scene is like a prefab or a blueprint, it can have a number of nodes (children components) and scenes (prefabs/blueprints) and can be instantiated. A collection of scenes is like a scene in Unity, or a map in UE4, it holds a bunch of prefabs/blueprints and represents a level, or a section of the game.
That last bit is a questionable design practice; I don't quite understand why a game's level needs to be treated as a special thing distinct from the regular prefab/blueprint system, as in other engines. Godot wisely eschews that and just doesn't treat the game's levels or maps or sections as being different from anything else. Makes the game much easier to compose IMO. A generic data-container for other generic data-containers is redundant, and creates pointless design problems like "should I make this a scene/map, or should I make this a prefab/blueprint?"
