Instead of storing members in two hashes - :members & :statics -
the members are now stored in simple :members array.

Lots of refactoring, but in the end the unit tests are all passing
and running JSDuck against ExtJS SDK works fine.  More work is needed
though as now JSDuck::Class contains several methods and algorithms
that are inappropriate in the context of new data structure.

Here I'm only changing the internals - export format is still the same
and so is the data structure given to generated Docs app.

This was an experiment, but turned out to be not very useful.
Better to leave the statistics generation to external tools.

Use 4 sections instead of previous 3:

- Main options
- Customizing output
- Tweaking
- Debugging

The last empty line is now added by JsDuck::OptionParser, so no
need to add it in every option description and inside
Logger#doc_warnings.

Additionally changed the default title to: "Documentation - JSDuck"

--help without arguments will now print a short list of all options.

--help=--option-name prints a detailed description of the given option.

Implemented with JsDuck::OptionParser class that extends builtin
OptionParser with the ability to print out help in two formats:
short list of all options & long description of specific option.

Treat this case as we do with other such comments in JavaScipt -
by defaulting to property.

The old Logger.fatal is now replaced with fatal_backtrace.

The actual patch is maintained in:

https://github.com/nene/esprima/tree/linenr-in-range

Having the esprima source right inside JSDuck source tree is more
convenient for other developers as there's no need to additionally clone
esprima repository in addition to jsduck itself.

Also the esprima.js needs to be modified only rarely.

This got broken by previous commit that added printing of filename.

Except for one case with @since tag where I can't access the context of
input filename.

Only in UNIX.

Because a class can consist of several files, list all the files
separated by spaces.

Using both Gemfile for bundler and a .gemspec for gem doesn't seem to
work quite well together.  Besides I'm not really making much use of
bunder, so I feel I'm better off without this additional complexity.

Though, this means I need to explicitly list the dependencies to be
installed for development, but there are only a few, so it really
shouldn't be much of an issue.

Allows defining a different version as a starting point for generating
@new tags.

Removed the old yellow "Private" sign from classes - they now have an
orange "private" bubble just like members.  All other meta tags now also
show up in class header.

Without this the lookup table will report duplicates that aren't really
there.

First off just a tag-based implementation of Ext4 overrides.