Simple helper to turn JsLiteralParser output back into string.

Helper for parsing JavaScript literals.

First parsing tokens into a buffer, so no token needs to be scanned
twice.  Still... the scanpointer of StringScanner is only advanced
when #next method is called.

The scanpointer of StringScanner is only advanced when Lexer#next called.

This allows us to do partial lexing on a string.

First off supporting String, Number, RegExp, and Boolean.

Supporting all JavaScript literals:

- Number
- String
- RegExp
- Boolean
- Array
- Object

Because the string "(optional)" is no more part of the description,
we need to add it when rendering the long parameters list.

Now supporting both:

    @param {Type} foo (optional)
    @param {Type} [foo]

Mostly just ensuring that we don't crash if bogus input is given.

The layout of example was a bit messed up when iframe loaded initially.
Tried waiting for Ext.onReady inside iframe - didn't help.
So currently just hard-coding a 100ms delay.
Crap.

Firefox doesn't support innerText (use innerHTML + stripTags).

According to the agreed-on spec for type-definitions-format with
the VJET Eclipse Plugin team.

Added bunch of tests to make it real solid.

Running it on SDK now produces some hords of warnings.

Previously it produced JsonP files - not that good for export.

Instead make use of the instance variables @opts and @relations.

Extracting a big chunk from the god class JsDuck::App.
The options object can now be easily passed around, unlike
with the previous system where every option needed to be
passed separately.

The bin/jsduck is now really short and readable.