1. Make text buffer dynamically allocated.
2. Rework scanner to get rid of *all* flex dummy rule fallbacks.
3. ID type attributes should be validated for uniqueness.
4. IDREF and IDREFS attributes should be validated.
5. ENTITY and ENTITIES attribute types should be supported.
6. The xml:spaces pragma attrbute should be supported.
7. Make the pcdata buffer act more like a traditional parser stack.
