Completely removed the unused sorting functionality.

Removed the @root variable.  Only local variable used inside create().
Generalized the logic in add_class and moved it entirely into create().

create() now returns array.  The surrounding hash is added inside App.

Renamed expandable_desc() to short_desc() - nothing expandable in it.

Few more small variable renames.

Replaced comments with more accurate ones.

JsDuck defaults to markdown anyway, so just ignore it.

In Ext4 codebase @author is quite frequent, so why not support it.

Because this class name was used as part of the cache key,
a class having method and event with same name would wrongly
use the cached version of method in place of event.

Now fixed.

The most basic support for mixins.  Detecting them when they are
listed in "mixins:" section at the beginning of Ext.define()
configuration object - some things can be between them, but it
will still break fairly easily.

- Listing mixed in classes in generated docs.
- Listing the mixed in members in methods, events, etc tables
  the same way as inherited members.

Support for the simplest case where "extend:" is the first item
in the config object passed to Ext.define.

Previously it was :assignment, which was a hack.

Now remembering source filename and line number of all doc-objects.

Strange how I forgot to add this really useful and trivial-to-implement feature.

Updated gemfile and documentation.

Even more speed!

To make this performance-boost work, some restructuring was required:

- Parser now includes DocParser, returning the doc-blocks
  in fully parsed form.

- Aggregator no more includes Parser.  Instead of parse() method
  it now has aggregate(), which expects output of Parser as input.
  Because of this some changes to unit tests were also needed.

- App performs the parallel parsing of source code and then
  feeds the results of parsing sequencially to Aggregator.

- Timing now reports three measures: parsing, aggregating, generating.
  But because of the parallel-phase, the report of parsing isn't
  that accurate as it also includes the generation of code files.

This gives us about 18% of speed improvement.

All tokens beginning with dash "/" are now handled in one place.

Minor speed gain.

skip_white_and_comments makes now less method calls.

This makes the generating of pages about twice as fast.

In verbose mode the time spent in parsing and generating gets printed.

Remove redundant "then"-s.

Let's be nice to those who prefer different coding style.

Now we have reached feature-parity with ext-doc.

Just ext-doc does it.

Updated README changelog also.

Using the same prettify.js that we use in source/.

For links inside code examples force showing of underlines -
otherwise prettify will make the links undistinguishable from
other source.