- Move JS/CSS parsing error handling up to the level where we read the
  file in, so when IO error occours we'll also catch that.

- Handle guide reading/formatting errors.

- Handle errors in formatting class docs (wasn't able to test it, but
  a user reported a stack trace with encoding error happening inside
  there - maybe a difference when running in Windows).

Previously when default value was 'false' it got ignored completely
as a result of just checking the truthiness of default value.  Now
explicitly checking that the value is not nil.

When parent class has config "foo" then the auto-detected property
"foo" in child class should also be a config.

Incorporated this logic in InheritDoc class.
Added separate test file.

It seems that VBulletin membergroupids have changed.  ID 7 no more
means a moderator - instead all moderators have 2 and dev team
has additional ID of 19.

The mapping of membergroupids to moderator status is now done just
once at login time.  Previously the group number was checked whenever
one wanted to determine moderator status, although a boolean moderator
field had been set at login.

Additionally fixed emailHash saving in ForumUser#clientUser - previously
the result of calculation was just thrown away.

Solved conflicts with @throws implementation and addition to CSS parser.

Not the prettiest solution, but at least it results in correct code
(unlike the previous concatenation without spaces).

Had to pretty much re-implement how --ext-namespaces works in new parser
context.  However, the implementation is much simpler than the one
for old parser.

So that by specifying --ext-namespaces=Foo.Bar we're able to recognize:

    Foo.Bar.define("MyClass", {
    });

CSS Specification uses the term <length> to denote 2em, 3px, etc
values with units.

It was only used by CssParser but that now uses CssLexer instead.

Using CSS lexer instead of JavaScript lexer simplifies CSS parsing
quite a bit (which is kind of obvious :)

It also removes some of the strange cases where CSS parsing failed
(like when the CSS identifier contained the word "default" which is
a keyword in JavaScript).

Happens quite often in Ext JS code where class doc-comment contains
@constructor and further down the code inside Ext.define() is the actual
constructor code that is auto-detected by JSDuck.

Unless it appears inside class doc-comment.

Generalize common auto-detection functionality to apply_autodetected
method, use it for detecting configs, statics, methods, properties.

Make use of the iterator method for looping over ObjectExpression.

Don't build a hash of keys and values, just iterate sequentially over
the ObjectExpression.  This will be needed when we next want to detect
all the normal methods and properties inside Ext.define().

Wrote the SourceFileParser#parse in different way.

The sort data had been changed so that it now contained meta field
instead of private and removed fields directly.

Fixed it, so that public members are now again listed before deprecated,
private and removed ones (as a new thing now also taking account the
deprecated ones).

Enhanced the documentation to further explain how the algorithm works.

Eliminated lots of magic numbers from code and replaced them with
variables - most of which are calculated from the three main constants
defined at the beginning.  This should make it much simpler to enhance
the function in future.