- Also updated the language for the code sections.

- Preliminary support with limitations on how the comments may be
  captured inside the nodes, as documented in the README.

- The example should have been included with the release... oh well, it
  will be in the next one (or on the git repo...)

- The example should have been included with the release... oh well, it
  will be in the next one (or on the git repo...)

- Readme updates to CI tags.
- CI files.
- Pre-release tag.

- Link with crimp, the CLI front-end.
- Note that this is now stable.

- Mention something about performance.
- Also noting the changes.

- Show that read/write streams work as advertised.

This required very substantial changes, some of which were cleanups that
were done on one stab included with this one, namely:

- Refactored the sourcemap.write function, the return value matches what
  documentation for the class said (mapping before names).
- Reordered the arguments for that, too - normalize is now the first
  keyword argument (should really be all keywords, but due to support
  for Python 2, this cannot be specified).

Now that write function and the walk function that feeds into it can be
both corrected in one go:

- Introduce a new ruletype for the final stream output - StreamFragment.
  This is a 5-tuple that includes the source, which will now be derived
  using the sourcepath attributes of input Nodes.
- However, in the interest of flexibility and not clobber the original
  logic completely, the unparser walk function now support a couple
  additional arguments, namely the conversion function which defaults to
  converting the TextChunk (formerly SourceChunk) type into the desired
  StreamFragment type so the source can be tracked.
- The other is a decorator function for the inner walk function which
  defaults to an implementation that will track the stack of Nodes that
  have sourcepath assigned, so that the source lookup can be done by
  checking the most recently pushed value without having to backtracked
  to its parent node for that value.  Not that parent references are
  available, however.
- This also allow the use of an identity function for both these
  arguments to return the walk function to produce the chunks without
  the source handling.  Alternatively provide a way to simply produce
  the text output only.  This is tested by the test_es5_walker module
  so that the SimpleChunk would not be turned into the StreamFragment
  which relies on attributes that it doesn't provide.
- To minimize disruption of existing tests, the test that checks the
  raw output of the Unparser, which is produced through the walk
  function, mostly will turn that back into a 4-tuple for checking
  against the existing values.
- Finally, for cases where nodes do not provide sourcepath, and given
  that the sourcemap requires a path (or url?) for the source, the
  about:invalid will be used (RFC6694, and the iana about-uri-tokens
  registry) to satisfy that value.

- Bring all the rule related setup functions into the core or rules
  module, where applicable.
- Formalize all rules constructor to become a callable, even though they
  may not have options, this makes invocation/usage consistent for end-
  users (even though the underlying classes point directly to the
  handlers themselves in the handlers.core module.

- Group all the layout handler related things into their own submodule,
- It's like how the parsers need lexers, unparsers need handlers.
- Indentator should be in its own module, like the Obfuscator; also
  the obfuscator module is renamed to obfuscation.
- Names should now better reflect the intended purposes.

- Yes, this was overkill, I am still assuming that I will get around to
  building some kind of es6/2016/2017/whatever support.

- Just too many problems otherwise.
- Also require a number of dumb compatibility fixes which may end up
  slowing some auxiliary operations (mostly operations involving repr).

- Since it was discarded already, accessing that value in the book will
  be the one from the previous chunk(s).  Simply use a 0 for that to
  have it be unchanged/inferred by the sourcemap algorithm.
- The extra vlq encoded tokens in the documentation should have clued in
  the mistake earlier given that they should have all been normalized
  like it is now.

- To make PyPI more happy.
- I do hate how they are making the process of making a release even
  more painful than it already was.

- So the breaking changes can be done to make life easier.

- Also, changed the indentation argument to indent on the PrettyPrinter
  class.

- Updated the background information and added a whole lot of other
  details.

- Also updated README to include the badges.