The code should be now both cleaner and faster (we're no more calling
look(/regex/) repeatedly to decide which at_tag method to call).

Remove bunch of duplication by parsing similar tags with a more
generic functions.

Also avoid repeating regexes by not testing for the exact tagname
any more after the first initial check.  The new helper method
parse_tag_as() replaces in lots of places two lines of code with
just one.

Suppose we're using JSDuck to generate guides from the following directory
structure:

somedir/
  guides.json
  guides/
    getting_started/
      README.md
      some-image.png
    nested/
      first_guide/
        README.md
        one.png
      second_guide/
        README.md
        one.png

In JSDuck's generated HTML, image tags in the Getting Started guide will point
to the correct path relative to the page (e.g.
'guides/guide_getting_started/some-image.png'). However, the nested guides will
specify the wrong paths to images: JSDuck will write
'guides/first_guide/one.png' instead of the correct
'guides/guide_nested_first_guide/one.png'.

This patch fixes the issue by generating the image's path based on the full
name of the guide, instead of using 'File.basename' to incorrectly crop the
path.

The strings can now contain the ] or } char without causing the
parsing to end at that point.  For example:

    @cfg {String} [foo="]"]

Also stars for classes with new members.

External classes can be now specified as patterns. For example
'Foo.*' to ignore all classes in Foo namespace.

It's now ignored together with all other possible tags within code
example blocks.

In Ruby 1.9 string[x] returns a string but in Ruby 1.8 it returns
a char code.  Fixed by using sting[x,y] which always returns a string.

One can now write doc-comment for event before the this.fireEvent()
and the event name will be picked up automatically, without any need
to add @event tag.

When encountering an unsupported @tag, JSDuck now prints a warning.

To achieve this, I've enhanced the doc-comment parser to only accept
@tags when they have a space in front of them.  Previously this would
have been detected as @method tag:

    /** foo@method.com */

Now it's treated as plain text as it should be. Only when separated
with spaces, the @method tag gets detected:

    /** foo @method .com */

Previously --ext-namespaces only applied to detecting Ext.define.

For backwards compatibility with old ExtJS sources that contain
@override to denote a method is overriding another one in parent.

One can now just add an empty doc-comment before Ext.override and
all the overrides will get detected:

    /** */
    Ext.override(Some.Class, {
        foo: 1,
	doSomething: function(){}
    });

That is, allow to use the @override tag without accompaning @class
tag to give a name for the override.

An off-by-one bug which prevented one from placing a comment right
before some other code construct without leaving a space in between.

This allows for a way to document classes that aren't created with
a standard Ext.define:

    createClass("Blah", /** @class Blah */ {
        foo: 17,
        doSometing: function() {},
        doOtherThing: function() {}
    });

All the members inside the object literal will end up inside the class.

Additionally this is equivalent to the functionality of @lends tag in
jsdoc-toolkit, but we achieve the same without having a separate tag.

The template.html, index-template.html and print-template.html now
all reference the same CSS files, allowing us to generate an hashed
app.css from it which in all three cases results in exactly the same
hash - so instead of three files we get just one CSS file.

Include prettify library also to concatenated CSS/JS.

Fix how JsDuck::TemplateDir copies over the files - ensuring that
it also copies the files containing MD5 hash in their names.

Now it can be also reused inside JSDuck itself.

Again, to keep the root of jsduck repo cleaner.

Now one can simply use these options instead of injecting special
comments-enabling JavaScript with --head-html.  The JavaScript side
is now also better as it accesses the commentsUrl and commentsDomain
properties of Docs.data object, not a lot more properties on the
root Docs object.

Instead of using the server-side PHP script, the browser detection
is done one the client side with a compact JavaScript snippet taken
from http://detectmobilebrowsers.com/

This allows the index page of the docs to be cached, lowering the
burden on servers.

Instead of crashing mysteriously, the circular @mixins and @extends
now result in fatal error message which clearly shows the
dependency chain that's causing the problem.