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/http.py

@@ -739,21 +739,29 @@ class HTTPClient:
         return self.request(Route('GET', '/oauth2/applications/@me'))
 
     @asyncio.coroutine
-    def get_gateway(self):
+    def get_gateway(self, *, encoding='json', v=6, zlib=True):
         try:
             data = yield from self.request(Route('GET', '/gateway'))
         except HTTPException as e:
             raise GatewayNotFound() from e
-        return data.get('url') + '?encoding=json&v=6'
+        if zlib:
+            value = '{0}?encoding={1}&v={2}&compress=zlib-stream'
+        else:
+            value = '{0}?encoding={1}&v={2}'
+        return value.format(data['url'], encoding, v)
 
     @asyncio.coroutine
-    def get_bot_gateway(self):
+    def get_bot_gateway(self, *, encoding='json', v=6, zlib=True):
         try:
             data = yield from self.request(Route('GET', '/gateway/bot'))
         except HTTPException as e:
             raise GatewayNotFound() from e
+
+        if zlib:
+            value = '{0}?encoding={1}&v={2}&compress=zlib-stream'
         else:
-            return data['shards'], data['url'] + '?encoding=json&v=6'
+            value = '{0}?encoding={1}&v={2}'
+        return data['shards'], value.format(data['url'], encoding, v)
 
     def get_user_info(self, user_id):
         return self.request(Route('GET', '/users/{user_id}', user_id=user_id))