One can now use {@link #static-foo} to link to static member #foo
in case there's instance member with the same name. This wasn't
previously possible.

Instead of adding two different sets of keys to one hash, build the
hash just by member name.  Look member name up from hash and filter
by type if needed.

Doing things in this more correct way brought up some bugs in other
places. Like {@link #foo-bar} being detected as link to member foo
of type bar, which happened to work. Now we only detect correct types.

Similarly the @inheritdoc happened to work because string interpolation
handles both symbols and strings equally well. Now it needs to be
converted to symbol explicitly.

All-in-all we now have more correct implementation.

Also turn off -link_ambiguous for Touch2.

Like when {@link #foo} refers to either #method-foo or #event-foo.

Also add new warning type: link_ambiguous.

-link now only hides warnings about links to non-existant items.

-link_private hides warnings about links to private members.

Default values make sense for properties too.  Previously I thought
that they don't but people are already using them, so here we go...

It simply makes sense to not require adding @method when it's obvious
without it that it's a comment for method.

All the attributes are now in this hash. That makes possible
further refactoring of the whole attributes listing creation
in method signatures.

The required attribute is detected in a different way than other
things inside attributes hash, but it fits there nicely.

Refactor those tags along with @protected and @template into
new property 'attributes'.

Remove bunch of duplicated code.

Improves performance a bit, as otherwise browser will keep animating
the invisible gif on the background.

Previously only word-characters were permitted, resulting in alias being
ignored.

In IE9 when file is opened from local filesystem the localStorage
property of window exists, but is undefined.

I guess meta-tags feature is now quite as solid as most of the other
stuff, while stats is a really new thing.

This should fix loading of meta-tags in Ruby 1.9 when path is
relative to current dir.  But we should do it anyway, as it's
done with all other paths to also support Windows.

Just for completeness.

Previously the whole contents of ~/template or ~/template-min
was copied over, which included files like README.md and build-js.html
that were only needed for development.

Now it's all a lot cleaner.

Previously the following:

    /** @cfg */
    timeout: 30 * 1000,

would have been detected having default value of 30.

Now such values are just ignored.

Instead of generating the background images with `secha slice`
just disabled them and defaulting to background color.

Looks much better now.

It should also work for IE, but haven't tested.

As all exports are in JSON format, it doesn't make sense to have
one export type called "json". The name "full" describes it better.

Renamed JsonExporter to FullExporter accordingly.

The Exporter performed two different tasks, which are now
better divided between two classes.

JsonPExporter is now AppExporter - the purpose of it is to
generate data for docs app, the actual format is decided by
ClassWriter.

Exporters now only deal with generation the data, ClassWriter
decides whether to write it into .json or .js files, or to STDOUT.

This is similar to json export format, with the difference that
only class and method names are included to exported JSON.

Added two classes:

- JsonExporter - exports single class to JSON format file.
- JsonPExporter - exports single class to JsonP format file in compact form.

Additionally one can now use: --output -
This will produce output to stdout.

Old --json    is now  --export=json
Old --stdout  is now  --export=json --output=-

The TemplateDir is now only responsible for copying or linking
the tempate directory.

It takes care of the creation of data.js file.

Didn't quite work out this way. Instead now separating out the
different tasks in output. For start the source code export is
now fully done inside SourceWriter class, which got a write_all
static method.

The main chunk of the normal HTML output is done by TemplateDir
class that at the moment does a bit too many things.

Splitting the output routines into three classes:

- StdoutExporter
- JsonExporter
- HtmlExporter

Additionally creating ClassWriter class for outputting the .js or .json
files. This one isn't pretty, but I have to share the output logic
between JsonExporter and StdoutExporter.

Another nastiness is the format_classes method which has to run BEFORE
writing out classes JSON, but AFTER the source HTML files have been
created - because of the parallel stuff. Brr...

At least the App class now has less responsibilities, but further
refactorings are needed.

Welcome, Guides, Videos, Examples and Categories now all have a
static #create method to be used instead of #new for instantiation.

The #create returns either the corresponding object instance or
NullObject which behaves as if it were the needed object, but
effectively does nothing.

Categories.create is a bit different - it first uses one of two
strategies for creating categories: FileCategories when filename
present or AutoCategories when not, then feeding the resulting
categories list to actual Categories object.