Improve the documentation of using vLLM. We now use vllm serve instead of using truss server.

[FredLiu876]

https://linear.app/baseten/issue/FDE-776/update-vllm-example-in-truss-examples

• Created two examples:

1. Using vllm serve, recommended for 99% of users. OpenAI compatible.
2. Existing example using truss server. No longer recommended, OpenAI compatible with slight modifications and use of Baseten bridge api. Only for custom inference.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}