another effc++ spurious warning?

[Gabriel Dos Reis]

Nathan Myers:

> We already have a feature to prevent warnings originating from system
> headers.  Why does it fail in this case?

I believe I have an idea about the answer.

<theory>
  This can happen only for template instantiations.

  When we instantiate templates, we create new nodes in the
  contexts of instantiations, those contexts are not in the
  system headers, but somewhere in user codes.  And we do not
  copy the bit saying that the decl is in a system header.
  Therefore the warnings get out.
</theory>

Someone has to check that theory.