Enable docker run arguments when starting the pod

[cblmemo]

Is your feature request related to a problem? Please describe.
Yes - we would like to use --entrypoint=/bin/bash to enable using image with non-default entrypoint as RunPod base image, like the official vLLM image vllm/vllm-openai:latest. That image comes with an entrypoint ["python3", "-m", "vllm.entrypoints.openai.api_server"] (reference here).

Describe the solution you'd like
Adding another argument docker_run_args in the create_pod API. That will be

Describe alternatives you've considered
Maybe just adds a boolean flag to override the entrypoint with /bin/bash - I guess it is common to use an image with proper dependency but a non-default entrypoint.

Additional context
Using vllm/vllm-openai:latest as base image gave me this error:

2024-08-29T23:42:51.938684641Z                      [--max-lora-rank MAX_LORA_RANK]
2024-08-29T23:42:51.938688657Z                      [--lora-extra-vocab-size LORA_EXTRA_VOCAB_SIZE]
2024-08-29T23:42:51.938692581Z                      [--lora-dtype {auto,float16,bfloat16,float32}]
2024-08-29T23:42:51.938696434Z                      (a lot of arguments here)
2024-08-29T23:42:51.938825141Z api_server.py: error: unrecognized arguments: bash -c echo xxx | base64 --decode > init.sh; bash init.sh

The script I use to create the pod is similar to #337, but only change the image_name arguments.