Refactored the guide filename and path detection a bit -
now also keeping a separate :dir field to reference the directory.
Using "url" field for that purpose seems awkward, so replaced
all these usages with :dir.

Fixes #584

Fixes #577

Specify sass version.

Because sass 3.3.x breaks some of the css related specs of jsduck.

Although the build on ruby 1.8.7 is still failing, this change fixes the builds on ruby 1.9.3 and 2.0.0 on Travis-ci.

Refs #536

A patch by Patrick Gleichmann.

Specs by me.

Refs #527

Fixes #515

Instead of relying on offset values, align the top-right of the menu
with bottom-right of the Docs app title.

Fixes #516

For IE8+ and other browsers the ExtJS 4 framework supplies a data-uri,
but for IE 6 and 7 it defaults to an URL pointing to sencha.com,
which will cause problems when offline.

Refs #511

Only do it for the generated TOC, but leave the original HTML alone.

Fixes #507

Implement a special '@browser' keyword, which when used as:

    --external=@browser

will include all the standard browser web API classes, listing
of which taken from:

https://developer.mozilla.org/en-US/docs/Web/API

Fixes #502

Instead of :external_classes, use :external - like the name of
the option itself, and like it is now with all other options.

Quite a hack... something better is surely needed.

Allows one to supply a list of unknown tag names on which the
warning is not shown.

Fixes #500

Change the #warn method so it can be used for also passing in
arguments, and then use it for reporting the :nodoc warning too.
For this to work:

- Logger#warn expects a `file` parameter instead of `filename` and
  `linenr`, which must always be a hash.
- The fourth parameter is then free to be used as optional `args`
  array for the warning types that need it.

The support for the deprecated old format was removed a while ago,
but the warning type was forgotten to be removed, so it hanged along
without having any actual effect.

JSDuck is really only meant for documenting JS and SCSS files.

Fixes #474

Avoid repeating option names as :foo and --foo.  Instead use a
function to transform from :foo to --foo.

Only leave the :title and calculate later the header from it in
Web::IndexHtml class.

Was broken since we implemented names for validators.

Always expect it to be an object with accessor methods.
As a default (and in tests) use OpenStruct.

So that the option name used inside JSDuck is the same as the --option
used on command line.

Rename the following:

* categories_path --> categories
* output_dir --> output
* template_dir --> template
* img_tpl --> img
* link_tpl --> link
* imports -->  import

Reset the defaults every time the Options::Parser#parse method is
called.  This allows us to re-use the parser which is slow to instantiate.

Some tests still need to re-instatiate the parser every time - move
these to a separate suite, so only the tests that need it will run
slower.

Implement full tests for them all.

Improve NullObject with the ability of values to be Proc's, which
will be invoked to provide more complicated logic.

Give names for all validators, so we can call them out by name in our
tests.

Add validator for :processes option.

Instead of calling optparser.on and opts.attribute directly, call
option and attribute methods that simply delegate to these objects.
This will also make it easier to refactor things in the future.

Eliminate the Validator class and attatch all the validation rules to
the Options::Record object, on which we can later simply call #validate!
to perform its own validations.

Previously the logic for this check was duplicated and sadly wrong.

So it hides all the internal options processing and we can use just
call this one static method in bin/jsduck.