Workaround for virtual function/dllimport bug, also patch guidance request

[Jason Merrill]

Do you think it's better to handle dll*port this way rather than the way MS
does?  We've been trying to use the same rules for both dll*port and
intra-shared object heuristics, which doesn't really make sense.

WRT virtuals, you need to wait to check DECL_VINDEX until the class is
complete.  But it looks like that could be fixed; it would just be a matter
of running check_for_override earlier, perhaps in grokclassfn.

Jason