TODOs:
- Documentation.
- Error handling for invalid type.
- Infer text stream by clustering.
- Support writing/creating PDFs.
- Any special treatments for linearized PDFs?