Late promise bindings would fire regardless of the outcome of the
promise.

Test case by: Jonas "zimbatm" Pfenniger <jonas@pfenniger.name>

It seems that the current __filename module global is mainly used to
determine the directory the current module is in. To make that
easier, this patch adds support for a __dirname module global
directly.

Because now, we expect the exception to be thrown if no errback
is given, we can't let timeout() add an errback silently.

Instead of JSON encoding them, just rethrow promise errors since that
produces much cleaner error messages.

A promise will throw an exception unless an error handler is attached in the
same "tick" that the error is emitted. This is to avoid silent promise
failures.

Listeners attached with addCallback / addErrback will now be executed
right away if the promise has already fired.

The current implementation was bad and nobody is using it. Has a chance
of getting re-implemented by somebody who has an actual need for it.

This is a replacement for the common hack:

  setTimeout(cb, 0);

It's much more efficient.

With priorities. Will be used for process.nextLoop().

For some reason v8 doesn't include the frame with the syntax error in the
stack trace - so have to special case it.

Why? Because I have two versions of GnuTLS installed - one is old 2.0.X in
/usr and one is new 2.5.X in ~/local/gnutls. waf correctly finds that the
newer version, but because GNUTLS was behind other libraries in the
node.uselib -L/usr/lib was before -L/home/ryan/local/gnutls/lib in the
actual gcc command - hence getting link errors. WAF SUCKS, really.
I wish someone would invent a good build system that could avoid such
problems.

eof_symbol wasn't initialized until the first packet arrived.
In this case no packets arrive before the client closes the socket - eof.

Also make normalizeArray a bit more proper wrt handling dots when keepBlanks
is enabled. Fix indentation.