M2wsgi

* Include REMOTE_ADDR in the WSGI environ dict; thanks steamraven.
* Add support for gevent 1.0b; thanks aadis.

* Compatability with latest eventlet and gevent releases.

* Compatability update; m2wsgi now requires Mongrel2 v1.6 or later.
* Removed the "reaper" device; Mongrel2 v1.6 has its own timeouts

* Add support for tnetstrings as well as JSON.
* Add WSGI "run_once" variable; thanks logic.

* A HUGE suite of API refactorings.
Too sweeping to report in detail, but the highlights are:
* abstract base classes now defined in m2wsgi.io.base, with concrete
implementations in m2wsgi.io.raw, m2wsgi.io.eventlet, etc.
* new class "Client" represents a client connection to which you can
send data; basically a (server id,conection id) tuple.
* new class "Handler" is a base class for handlers, with nothing
WSGI-specific. Maybe useful for writing custom handlers.
* different types of Connection now live in different classes, rather
than a single class with various socket-type switches.
* see m2wsgi.io.base.DispatcherConnection for an example.
* Socket identities are now specified as part of a single socket-spec
string, with a URL-like format, e.g: tcp://IDENTITYhost:port
* New "dispatcher" device replaces the old pull2xreq device:
* sends explicit pings to detect dead handler processes
* capable of flexible request routing
* supports "sticky sessions" out of the box
* New "response" device gives out easy canned responses.
* New "reaper" device implements simplistic request timeouts.
* added some utility middleware for use by the server:
* GZipMiddleware: gzip response bodies
* BufferMiddleware: buffer response bodies, for better compression

* New method Connection.shutdown() to stop receiving more requests,
but allow processing of any that are already received.
* Much improved custom request protocol with clean shutdown:
* socket type changed from from REQ to XREQ.
* helper renamed from pull2queue => devices.pull2xreq
* handler now sends an explicit shutdown message to the socket.
* socket can push requests to the handler asynchronously.
* WSGIHandler._shutdown() now waits for all in-progress requests to
complete before it terminates the server.
* Add experimental gevent IO module
* Install command-line script "m2wsgi".