config steps for serving via public-ip

[chauhang]

The default configuration works for serving models on localhost. In order to run predictions on models via the public-ip (making calls from a different host) one has to specify 0.0.0.0 or the explicit IP-address through the --ts-config file settings. This is not very clear in the documentation. Add clear instructions for new users to get started, otherwise can take a while to figure out why predictions are not working even after opening up the external security ports on the machine.

[mycpuorg]

3e98872 has this. Let me know.