CommonMark 0.24

[0.24]

• New format for spec tests, new lua formatter for specs.
  The format for the spec examples has changed from

  .
  markdown
  .
  html
  .
  

  to

  ```````````````````````````````` example
  markdown
  .
  html
  ````````````````````````````````
  

  One advantage of this is that spec.txt becomes a valid
  CommonMark file.

• Change tests/spec_test.py to use the new format.

• Replace tools/makespec.py with a lua script, tools/make_spec.lua,
  which uses the lcmark rock (and indirectly libcmark). It can
  generate HTML, LaTeX, and CommonMark versions of the spec. Pandoc
  is no longer needed for the latex/PDF version. And, since the new
  program uses the cmark API and operates directly on the parse tree,
  we get much more reliable translations than we got with the old
  Python script (#387).

• Remove whitelist of valid schemes. Now a scheme is any sequence
  of 2-32 characters, beginning with an ASCII letter, and containing
  only ASCII letters, digits, and the symbols -, +, ..
  Discussion at http://talk.commonmark.org/t/555.

• Added an example: URI schemes must be more than one character.

• Disallow spaces in link destinations, even inside pointy braces.
  Discussion at http://talk.commonmark.org/t/779 and
  http://talk.commonmark.org/t/1287/12.

• Modify setext heading spec to allow multiline headings.
  Text like

  Foo
  bar
  ---
  baz
  

  is now interpreted as heading + paragraph, rather than
  paragraph + thematic break + paragraph.

• Call U+FFFD the REPLACEMENT CHARACTER, not the "unknown code
  point character."

• Change misleading undefined entity name example.

• Remove misleading claim about entity references in raw HTML
  (a regression in 0.23). Entity references are not treated
  as literal text in raw HTML; they are just passed through.

• CommonMark.dtd: allow item in custom_block.