`Define` now returns the same node it is called on rather than a new
node of the functional graph.
This is a performance optimization as it removes the need to traverse
these nodes during the event loop, e.g. when checking filters in a
functional chain.

Behaviour is unchanged (except for `Define`s return type, which should
never be explicitly typed out by users anyway) and users can keep
treating `Define` _as if_ it returned a new node of the functional
graph.