This fixes JSDuck crash when keywords used as property names.

URL-encode all unicode characters.

Fixes #397

After Process::InheritDoc has run we regenerate the ID-s of all
members and invalidate all member indexes.  The old code tried
to be clever and only recalculate members that had changed and
classes containing these, but it failed to do so correctly,
resulting in this hard-to-find bug.  Now I've taken a slower
approach of just recalculating everything, which should now
give us correct results.

Fixes #345

Only read the (required) flag for the config itself, but for the
subproperties leave it inside the text.

Fixes #406

Using eval() was a hack and caused a string like "#$%" to crash
JSDuck as Ruby tried to access the global $% variable.

Fixes #419

Detect property name and type when doc-comment followed by
Object.defineProperty().

Fixes #241

When navigating to a class member or guide section.

To undo the effect of horizontal scrolling that may have been
done before to view an image that didn't fit in width.

Fixes #411

Don't crash when encountering [5, , 7]

Although nobody shouldn't write such code in their sane mind, it's
still valid JavaScript.

Fixes #418

The latest RKelly Remix provides a hackless way to access this token.

Related to #415

A hacky solution for start as the RKelly::Parser doesn't allow
public access to the token at which it stopped.

Improvement of #415

Instead of crashing with a mysterious "undefined method `range`
for nil:NilClass" message, raise our own exception, telling that
the problem is with JavaScript syntax.

Fixes #415

Use simple File glob patterns to include only the files we want,
not having to exclude files we don't want.

Reference all the latest versions available currently.

This will avoid breaking of JSDuck when one of the dependencies
has been upgraded to become imcompatible.  But it also introduces
a responsibility for us to update these version numbers whenever
a new JSDuck gem is released.

Fixes #408

Fixes: #405

Use the ~> syntax to allow us to release patches to rkelly-remix
and get the picked up by jsduck installations.

When for-in contains variable declaration, Esprima produced the
whole VariableDeclaration, but RKelly gives just VarDeclNode -
so we wrap it inside a VariabelDeclaration to match Esprima output.

Missing variable in catch clause is a JS syntax error.
This happened to work in old Esprima, but no more works in
neither new Esprima nor RKelly.

That's not supported by RKelly, and really not needed in practice.

It still referenced the old Esprima engine.

Because the EsprimaAdapter gives a string value to regex literals,
which is different from nil received from Esprima itself, we'll
have to check first for regex and after that for string.

Previously && and || were treated as BinaryExpressions, but Esprima
classified these as LogicalExpressions.

The "test" and "body" parts had been switched.

So we don't have to change all the tests that target the old Js::Parser.
Besides, it's better to use a more generic name anyway.

Break the no_doc warning type into three:

- no_doc
- no_doc_member
- no_doc_param

At least in browsers that support it.
Unfortunately Chrome doesn't.

Previously H5 and H6 weren't styled at all, resulting them looking
like plain text.