Add SageMaker vLLM Runbook: Endpoint Creation, Handler Override, Pre/Post Processing

[HappyAmazonian]

This PR adds runbook documentation and working examples for deploying vLLM models on Amazon SageMaker, covering three key topics:

1. Endpoint Creation - Basic deployment workflow
2. Handler Override - Customizing /ping and /invocations endpoints
3. Pre/Post Processing - Adding middleware for request/response transformation

What's Included

Documentation (docs/sagemaker/)

• 01_quickstart.md - Endpoint creation guide
• 02_customize_handlers.md - Handler override guide
• 03_customize_pre_post_processing.md - Pre/post processing middleware guide

Jupyter Notebooks (examples/vllm/notebooks/)

• basic_endpoint.ipynb - End-to-end endpoint creation example
• copy_image.ipynb - Copy vLLM container images to your ECR
• handler_customization_methods.ipynb - Handler override examples
• preprocessing_postprocessing_methods.ipynb - Pre/post processing examples

Model Artifact Examples (examples/vllm/model_artifacts_examples/)

• handlers_decorator.py - Handler override using decorators
• handlers_discovery.py - Handler override using function discovery
• handlers_env_var.py - Handler override using environment variables
• preprocessing_postprocessing.py - Pre/post processing middleware examples

[Lokiiiiii]

Why is this required ? Can the public ECRs not be directly used with SageMaker ?

[HappyAmazonian]

Yes. It is prohibited by the hosting team. Customer either needs to upload the image or modify VPC setting

[Lokiiiiii]

Merging this since we don't have a workaround for now. Please take an action item to update this once this issue is fixed.