fix: do not reset the ping timer after upgrade

There was two issues with this behavior:

- v3 clients (with allowEIO3: true) were also receiving a "ping" after
a successful upgrade, which is incorrect (in v3, it's the client that
sends the "ping", and the server answers with a "pong")

- the ping timer is not reset after upgrade on the client-side, so an
upgrade which took longer than the `pingTimeout` duration could lead to
a "ping timeout" error on the client-side

I think the latter issue is present since the initial implementation.

Related: socketio/socket.io-client-swift#1309 (comment)

Backported from ff2b8ab

lib/socket.js

@@ -208,7 +208,6 @@ Socket.prototype.maybeUpgrade = function (transport) {
       self.clearTransport();
       self.setTransport(transport);
       self.emit('upgrade', transport);
-      self.setPingTimeout();
       self.flush();
       if (self.readyState === 'closing') {
         transport.close(function () {

test/server.js

@@ -1092,6 +1092,24 @@ describe('server', function () {
       });
     });
 
+    it('should not timeout after an upgrade', done => {
+      const opts = { pingInterval: 200, pingTimeout: 20 };
+      const engine = listen(opts, port => {
+        const socket = new eioc.Socket('ws://localhost:%d'.s(port));
+        socket.on('open', () => {
+          setTimeout(() => {
+            socket.removeListener('close');
+            engine.close();
+            socket.close();
+            done();
+          }, 500);
+        });
+        socket.on('close', () => {
+          done(new Error('should not happen'));
+        });
+      });
+    });
+
     it('should not crash when messing with Object prototype', function (done) {
       Object.prototype.foo = 'bar'; // eslint-disable-line no-extend-native
       var engine = listen({ allowUpgrades: true }, function (port) {