Strip external generics on build

[hadley]

Fixes #364

[lawremi]

Might want to clarify in the docs that it should come after all method registration. The example in zzz.R shows that, but it might help to be explicit.

This also raises the idea of a combined function S7::load() that clears generics and defines .onLoad() via side-effect. Probably not a very good idea, because confusion could result from another .onLoad() being defined. I wonder if the namespace system could support .onLoad.<foo> so that there can be multiple hooks.

[hadley]

@lawremi Claude discovered this while writing this test. Does that seem correct to you? (i.e. that S4_register(), which calls setOldClass() has to be called onLoad)

[hadley]

Or maybe this is related to the environment fixes in #643?

[lawremi]

Shouldn't be necessary to call S4_register() at runtime. All it does is call setOldClass() which of course works at build time. Strangely, Gemini suggested the same thing when converting a package to use S7.

[hadley]

Let's hope that #663 fixes this. I'll hold off on merging until that's confirmed.

[hadley]

Hmmm, I wonder if we can call setHook(packageEvent(current_package, "onLoad")) to register an onLoad hook in S7_on_build()?

...

Nope, that doesn't work.