Segfault was due to `glyph_position` structs keeping a pointer to
`glyph_info` objects which went out of scope at a different time.
The (rather ugly) fix for the moment is to copy that information
into the thunk object.
This is done by creating a fake 'virtual' environment at a fake
point and running the symbolizer render code. The actual render
is saved in a thunk for after the group layout has been done.
This involved making the freetype engine stack allocated and
accessible via a `shared_ptr` because `FT_Face` elements stored
in it are referenced from cairo structures which persist beyond
the lifetime of the `cairo_renderer` object.
Hopefully this can be resolved in a future commit.
The idea here is that much of the code in the renderers and the
various `process_*` rendering methods is shared, but this sharing
is obfuscated by the lack of sharing of renderer fields and
methods.
This is the first step in trying to factor out commonalities in
order to simplify the implementation of the renderers and make
them easier to understand and modify.
