Ensure Readiness Probe Accuracy for devserver

[DevineLiu]

eg：

func main() {
	flag.Parse()
	var c config.Config
	conf.MustLoad(*configFile, &c)
	// When executing MustNewServer, the dev HTTP server starts
	server := rest.MustNewServer(c.RestConf)
	defer server.Stop()
	// Sleeping for 100 seconds is used to simulate scenarios like loading cache data
	// During this 100-second sleep, if you curl http://xx.xxx.xx:xxx/ready, it will return 200 OK
	// However, at this point, the REST server is not fully ready,
	// as health.AddProbe(healthManager) has not been executed
	time.Sleep(100 * time.Second)
	ctx := svc.NewServiceContext(&c)
	handler.RegisterHandlers(server, ctx)
	fmt.Printf("Starting server at %s:%d...\n", c.Host, c.Port)
	// health.AddProbe(healthManager) is executed only in start()
	// I think health.AddProbe(healthManager) should be moved into NewServer
	server.Start()
}

I think health.AddProbe should be added in NewServer, and then the server should be marked as ready in Start.
Otherwise, the /ping endpoint of the devserver will be ready even when the rest service is not ready.

A typical case is when we use the devserver /ping endpoint as the readiness probe for Kubernetes pods.
If the rest server is not actually ready, traffic will be routed to the service prematurely.

ref:

go-zero/zrpc/internal/rpcserver.go

Line 66 in 4ac8b49

health.AddProbe(s.healthManager)

go-zero/rest/internal/starter.go

Line 59 in 4ac8b49

health.AddProbe(healthManager)