Let the #hw method return it's original return value, which is the
DocScanner class - then it chained with one of the DocScanner
methods like #look or #match.  The only exception being #default_value
method which belongs to StandardTagParser and can't be therefore
chained from #hw method - but that's just a one little exception
we can live with.

The parsing now takes a config parameter to turn on/off parsing of Type
and Name sections.

This cleans up the DocScanner class considerably as all those complex
methods are now in StandardTagParser, leaving the DocScanner with
basics such as #look and #match, plus the big #standard_tag.

For now, it parses the following pattern:

    @tag {Type} [some.name=default]

That's the only place it's used in.

These are no more called by any Tag classes.

When somebody needs to parse \w+ pattern - then it should tell so
explicitly, but #ident will now parse the actual identifier pattern.

The member_reference parsing has now moved completely inside Inheritdoc
class.

So the #parse method of Tag classes doesn't have to return this
empty field and maybe in the future we get rid of it altogether.

Also introduce shorthand for horizontal whitespace skipping.

This will then be added as a new tag through #add_tag.

Extract DocMap class for building map of @tags in comments.  Run
the map building just once inside FileParser - instead of repeating
this thing in DocType and DocAst.

No real speed advantage, mainly removal of duplication.

Instead of performing the map building in every #create_* method,
the map is passed as parameter to all of them.

The tagnames from doc-comments and final field name are now the same -
:aliases.  Instead of the former case when one was named :alias and the
other :aliases.

Create ClassListTag base class for processing four tags:

- @mixins
- @alternateClassName
- @requires
- @uses

For this to work out correctly, the application of default values
in merger needed to be changes so that defaults get applied even
when the field doesn't exist in neither docs or code hash.

That is for tags: @cfg, @property, @method, @event, @css_mixin,
@css_var.  For the latter introduced also a Tag class (although
there is no corresponding @tag).

The regex for matching member types now gets constructed in TagRegistry
class, which has the central knowledge of the available member types.
Similarly exporting uses TagRegistry.member_types to loop over all
available member types (removing hardcoded member types list from
JsDuck::Class#each_member_type).

Otherwise the documentation inheriting process will run into infinite
loop.

For example the following code cause a problem:

    /**
     * @class
     * @extends B
     */
    A = {};

    /**
     * @class
     * @extends A
     */
    B = {c: 'd'};

That's fixed now.

No more does using this option restrict viewing of inline examples
to WebKit based browsers.  This check is now moved over to
eg-iframe.html in Sencha Touch codebase.