docs: replace stateful ggproto example with stateless one

[CuiweiG]

Closes #6582.

The current ?ggproto example demonstrates a stateful class (Adder$x) that, per the issue, "advertises a property of ggproto classes that should be used sparingly and ideally not at all if you're developing extensions."

The replacement builds a stateless Math class with two static methods plus a Mather child overriding one method via ggproto_parent(). Construction, method dispatch, and parent-method invocation are all still covered; no self$* field access appears anywhere. The child method carries self in its signature solely to thread identity into ggproto_parent(Math, self).

The "Working with ggproto classes" section gains a short paragraph stating that most ggplot2 layer classes (Geoms, Stats, Scales) are stateless, and encouraging extension authors to follow that convention.

Local verification: all example code paths run under devtools::load_all() with the expected values.

[teunbrand]

Thanks for preparing a PR!
I think the proposed change is not exactly what we want users to learn about ggproto classes. It may have very well be my insufficient explanation in #6582 that had set this on the wrong track.
To explain a bit more; it is fine to read from a ggproto object's fields, but to write to that field is discouraged.
In the old example, self$x <- self$x + n is the problematic piece.

[CuiweiG]

Thanks for the clarification — the write-vs-read distinction wasn't clear to me from the issue alone. Will revise the example to keep the field-based inheritance pattern but drop the mutation. Back to you shortly.