* Fixed indetation and spacing.
* Removed unused rules.
* Combined duplicated rules.
* Prefixed most stuff with .body-wrap class.
* Ordered things using some logic.

The source comes from official ExtJS documentation (included in
ExtJS release package).  This includes the minified version of
ExtJS 3.3.1 with gray style.

The original ext-3.3.1/docs/resouces contained a lot of unused
CSS and image files - these were all excluded as unnecessary crap.

JsDuck now includes --template command-line-option to specify
alternate template to use.  By default the template/ dir included
with jsduck will be used.

Added TreeIcons class that extracts icons from tree.

RDiscount is way faster and it doesn't throw errors for malformed
HTML like Maruku - so we can treat all the comments as Markdown,
plus we can mix HTML and Markdown.

Actually RDiscount doesn't throw any errors at all, whatever we
feed it.  If it doesn't recognize text as correct Markdown, it
doesn't complain - it just leaves it as is, which is perfect for
us.

There are some minor issues still with it:

- <pre><code>-s getting nested when parsing Ext source.
- Even one-liners are wrapped inside <p>...</p>.

But these are pretty easily solvable.

This way we can later style the markdown-HTML differently.

It does now more than just links replacement.

Expanded DocLinks to also perform Markdown formatting - but only
when doc-comment doesn't already contain HTML.  Event then we
might get error when parsing it as Markdown - in that case we
silence down the error and continue without rendering it as
Markdown.

Plus some ExtJS strangenesses in use of @member.

Reverse-engineered the behaviour of ext-doc when merging multiple
doc-blocks for the same class.  Not 100% the same, but mostly.

Now finally we have support for most of the stuff that ext-doc does
on parsing-side.  The output-side needs quite a bit more effort,
but parsing and analyzing of input seems almost complete now.

This siplifies tests that only need to include the class
they want to test and not worry about including all the
dependencies too, as these will be automatically included
by the class itself.

Additionally the header of each file gives a nice overview
of the direct dependencies each class has.

I copied the previous lump-all-together-into-main-file-design
from some ruby project in github, thinking there is some value
in doing so.  Well there wasn't...

This way all the @member definitions that refer to classes in
separate files will be nicely resolved.

JsDuck::Class no more rewrites :member attribute of its members -
it doesn't even need to know that there is such attribute.
While JsDuck::Table can assume that each class member still has
the :member attribute set.

DocParser now recognizes @member and Merger sets it for class
members if it finds it.

A lot of changes to Aggregator to support it.  Also added rspec
to check that it all works as intended.

Also tuned the expandability detection algorithm.  Now all items
with extra_doc are expandable, and the short version never contains
extra_doc.  Plus the ellipsis is only shown when the primary_doc
is shortened.

As there is next going to be LongParams.

Removed parameter list creation code from Table class as it's
only needed for EventTable and MethodTable - so it's now in
separate class that only those table-classes use.

The rest of the signature is pretty standard.

Currently ignoring the link href value in signature -
as we currently don't output source files.

Each documentation table type is now created by separate class.
Common functionality being defined in baseclass JsDuck::Table.

But still avoiding "stupid expansions".

Stripping HTML tags from short versions of doc-rows.