Previously only the first "." got replaced in classname.

Before: Ext-panel.Panel-cfg-baseCls
After:  Ext-panel-Panel-cfg-baseCls

Previously the ID-part of members table rows was also cached,
resulting in ID containing the name of the class that happened to
be the first for which the row was generated.

That will contain list of mixins from the class and all its parent
classes.

Previously --img and --link only applied to export, but now they
can also be used to override the links/images in the normal output
of JSDuck.  Currently it's most useful for dealing with {@img} tags
in ExtJS 4.

Instead of linking to line numbers, documentable items in source files
are now nicely marked up, and are linked by name.  This should prevent
links to source from becoming outdated too quickly.

before: Ext.html#line-86
after:  Ext.html#Ext-method-apply

All the output is now done after parsing etc has finished.

All parsing and merging now happens inside that class.
Consequently the Agggregator is now a lot simpler, as
it takes for input just instance of SourceFile.

SourceFile instances contain all the input: filename, source, and
parsed doc-objects.

The output of HTML source files now happens after all the
parsing and aggregating, similarly to all other output.
For now the links to source code are still line-nr based,
but this will now be easy to change.

These allow to define templates to use in export for replacing
{@link} and {@img} tags.

Adjusting the markup of links and images is now much more flexible.
This will be most useful for customizing the markup in exports.

The tests also got simpler, as we can test agains simpler markup now.

The name is now in plural (although the singular @tag works too,
but inside JSDuck the name is now "alternateClassNames" and is also
exported with that name).

Overall it behaves like @mixins tag, which too can contain multiple
items.

Correctly implemented merging of mixins and alternateClassNames when
merging class definitions.

Alternate classnames are also shown in JSDuck generated docs.

When @mixins tag present, it will override all the implied mixins
possibly detected from code.  One can either use just one @mixins:

    @mixins Foo Bar Baz

Or several:

    @mixins Foo
    @mixins Bar
    @mixins Baz

The result will be the same.

So that the e-mail address after author name will also be shown.

Hard to get it correct when there are:

- Ext.extend(...
- Ext.define("MyClass", {extend: "Parent", ...

But in doc-comment you had to previously write "@extends".
Now it's similar to @returns, which can also be @return.

No matter if alternateClassName is missing, is a string or an array,
the result of parsing will always be an array - even if just one
item or empty.

Simplified the parsing - using array_of_strings method to do most
of the work.

Doing a proper parsing now, so automatic link-creation won't be
done inside {@link} and {@img}.

Simple-minded implementation with regex search-replace.

When parameter or return type matches a classname included to
docs, a link to it is created.

Previously the search by class name almost worked, as we searched
also from constructor names, which happen to be the same as class
name.  Constructors are now excluded - little use of searching
from them.

Search now more intelligently looks first for exact matches,
then matches at the beginning and finally matches at middle.
This way the best matches are at the top.
Maximum of 50 results is returned.

DocFormatter will now look for names like Ext.form.Field and convert
them automatically to links.  The same works for method names, when
written like MyClass#method.

Only links to existing classes are created.  So that when there is
Foo.Bar in method description but the Foo.Bar class doesn't exists
(not included to documentation), then no link is created.

Additionally a DocFormatter instance is now created only once per
the creation of one documentation page, and then passed on to all
the classes taking part of page creation.  So, no more repeated
initialization code.

Now all the evals have been stripped from Lexer.
Bad things they were.

Variables, functions, etc not assigned to any class are listed there.

This class only gets created if there is at least one member to place
into it.

DocParser constructor now takes mode parameter that sets either
JavaScript identifier or CSS identifier pattern.

Less confusion this way with JavaScript mixin classes.

When @var found with @member My.Class, but no @class My.Class found,
then My.Class gets created.

Implemented basic parsing of CSS variable doc-comments.