* Fully implement skipping "junk" lines.
  * Mostly done now, but not rigorously tested.
* (Optionally?) remove the '' key (which stores the entire matched text) from 
  result hashes.
* Investigate segfaults in global destruction after parsing.
* Resync with spec
  * Handle \# in test descriptions, when the spec is updated for it
  * Spaces between # and directive, and directive and reason, should be optional
  * Clarify handling of unknown directives
* Comprehensive test suite.
