Reimplement zlib streaming.

This time with less bugs. It turned out that the crash was due to a
synchronisation issue between the pending reads and the actual shard
polling mechanism.

Essentially the pending reads would be cancelled via a simple bool but
there would still be a pass left and thus we would have a single
pending read left before or after running the polling mechanism and
this would cause a race condition.

Now the pending read mechanism is properly waited for before returning
control back to the caller.

discord/gateway.py

@@ -186,6 +186,8 @@ class DiscordWebSocket(websockets.client.WebSocketClientProtocol):
         # ws related stuff
         self.session_id = None
         self.sequence = None
+        self._zlib = zlib.decompressobj()
+        self._buffer = bytearray()
 
     @classmethod
     @asyncio.coroutine
@@ -312,8 +314,17 @@ class DiscordWebSocket(websockets.client.WebSocketClientProtocol):
         self._dispatch('socket_raw_receive', msg)
 
         if isinstance(msg, bytes):
-            msg = zlib.decompress(msg, 15, 10490000) # This is 10 MiB
-            msg = msg.decode('utf-8')
+            self._buffer.extend(msg)
+
+            if len(msg) >= 4:
+                if msg[-4:] == b'\x00\x00\xff\xff':
+                    msg = self._zlib.decompress(self._buffer)
+                    msg = msg.decode('utf-8')
+                    self._buffer = bytearray()
+                else:
+                    return
+            else:
+                return
 
         msg = json.loads(msg)