The class was still using the old JsonDuck instead of Util::Json.

Split the huge method into lots of smaller routines.

Move all the messy parsing logic away from already overcomplicated App
class.  Now the App is left with pretty much just one method that
represents the flow of the entire app.  It's still too complicated
though.

This new class encapsulates the formatting of all classes.

Also, while formatting, we now create a new class_formatter for each
class that we process.  That's because otherwise the list of images
detected in one class will be added to the images detected from the next
one, and so on and on...  This wasn't really a problem because we
stripped away the duplicate images anyway, but it wasn't really the
right way to do it.

- Constructor now takes just opts and relations is passed in separately
  through accessor.
- Remove old leftover: @images variable.

No reason at all for it to be part of DocFormatter.

Cancel out the min-width set in stylesheet.
Add <meta> to use device width as default.

Although our simple implementation worked also just fine, the builtin
Singleton has some more checks built into it - just good to rely on
something that's well tried and tested.

Using PHP Mobile_Detect library to do the device detection:
http://code.google.com/p/php-mobile-detect/

This mainly changes how we serve the front page - otherwise mobile is
treated exactly the same as print pages.

So that when the docs are loaded over file: protocol we'll still load
the fonts over http:, not attempting to load them from local filesystem.

Caused by the fact that IE8 doesn't allow the reserved keyword 'enum' to
be used inside a member expression, like so:

    someObj.enum

So I had to rewrite it like so:

    someObj["enum"]

IE9 and other browsers didn't have a problem with that.

Enhance Util::Json.parse to accept an options parameter which Esprima
class needs to pass.  Now everything accesses the JSON gem through
JsDuck::Util::Json wrapper.

Results in much simpler code without all the self.method_name uses.

Use it to abstract away the singleton logic in Logger.

This way splitting the assets into subgroups will help to reduce the
amount of information shown in ThumbList display.

Nested groups can be defined as follows:

    [{
        "title": "Main group",
        "items": [
            {"title: "subgroup",
             "items: [ ... ]}
        ]
    }]

In UI the tree will then show subfolders for the nested asset groups.
But the ThumbList display will still show all the assets in old way,
and the assets in subgroups will be expanded so they will appear as if
they were listed directly inside the top level group.

I guess in some way this beats the whole point of having the subgroups -
as all the assets still show up on the main page as if there were no
subgroups at all.  On the other hand we now technically have support for
subgroups, but I feel that's not really what these users were after when
they asked for nested grouping.

Additionally eliminated :dup_asset warning.  It didn't fit nicely to the
new way of iterating over assets, and I think it wasn't such a useful
warning anyway.

The class was for long no more in use, but the file was left undeleted.

A bit less stuff in the main JsDuck namespace.

Included a mini-library 'regexp-quote' which provides just one but
useful function.

Esprima now rejects those pieces of code as invalid JavaScript.

The new esprima no more considers newlines as part of line comments.

Let Esprima have its window object - otherwise it'll just crash.