GIOMM_CFLAGS got commented out, so we have to use GLIBMM_CFLAGS for
now.

I treated "continue" as a fall through. Seems that it does not work
that way.

Add H_INCLUDE section - used by GError wrappers for including wrap
init header and by GTyped enums/flags for including value
header. Added own public and private header inclusion in
implementation files.

These can be used as gmmproc-only comments, that won't be pasted into
generated source.

Because of enum breaking the scoping rule, its values has to be unique
in outer scope. So, for now we can't use names from gir files - we
have to use full names without global prefix. There is an option to
use new style enums, that is - enum classes from C++11, which we might
want to use in future. In new style enums, we can just use names from
gir without fear of multiple definitions of the same name.

Copy everything between angle brackets into output. Especially, we
don't want to parse class keyword that may happen there.

It is provided mainly for consistency - it simply calls gobj_copy()
method.

It is not provided in gir anyway. If we want to provide it then we
will have to add a way to inject enum values into _WRAP_ENUM or
_WRAP_GERROR.

We are not wrapping it anyway. But if we want to do it, then we will
have to provide some kind of override or addition for some API wrapped
there - some of it is not available in gir files.

It was exposed by g_bytes_{hash,equal,compare}.

For now gmmproc warns when there is missing transfer annotation or
constructor is incorrectly parsed as a method.

* .gitignore: Ignore bytes.{cc,h}.

WrapParser has no nl function - the function using it should be moved
to Output directory anyway (as TODO says).

Also, removed lots of files.