feat(runner.sh): only enable prefix caching and disable log request

feat(runner.sh): --enable-chunked-prefill and --enable-prefix-caching for faster generate

fix(runner.sh): enable eager mode (disabling cuda graph)

fix(runner.sh): --enforce-eager not support values

fix(runner.sh): explicitly disabling enforce_eager

fix(runner.sh): disable eager-loading so it using cuda graph (in order for parallel and faster processing)

feat(runner.sh): add specific task and code revision

feat(runner.sh): using MODEL_ID only

feat(runner.sh): using runner.sh to select llm in the run time