TODOs:
  - Documentation.
  - Error handling for invalid type.
  - Infer text stream by clustering.

  - Support writing/creating PDFs.
  - Any special treatments for linearized PDFs?