Fix unhandled non-field struct children

[serenity4]

Unhandled non-field child nodes were sometimes present in children(cursor), and were already handled in printing but not in codegen, which triggered an assertion error at

Clang.jl/src/generator/print.jl

Line 119 in 30aaa5d

@assert length(child_nodes) == length(fields)

where fields built from the codegened expression had an extra field for a forward struct declaration.

Here is an exemple header where the error appeared:

typedef struct _ExampleStruct {
  struct _ForwardA *a;
} ExampleStruct;

typedef struct _ForwardA {
  int a;
} ForwardA;

and codegen produced something like

struct _ExampleStruct
  _ForwardA::_ForwardA
  a::Ptr{_ForwardA}
end

Now we don't have that weird spurious _ForwardA::_ForwardA field.

It happened to me while wrapping a larger header (winuser.h) using this pattern, but I couldn't come up with a minimal reproducer. For some reason, fields(Clang.getCursorType(cursor)) for _ExampleStruct was empty in that larger header and Clang.jl fell back to using children to extract fields, while in any example I have tried for a minimal reproducer fields(Clang.getCursorType(cusor)) succeeded in returning the only field present as expected.

[Gnimuc]

_emit_getproperty_ptr! may also need to be patched.

[Gnimuc]

Could be related to the upstream changes discussed in #472

[Gnimuc]

let's wait for more bug reports.