Add AWS Deadline Cloud to Hugging Face on AWS docs

[anika-suman-amazon]

What does this PR do?

Adds AWS Deadline Cloud alongside SageMaker, Bedrock, Inference Endpoints, and EC2/ECS/EKS as an option for running Hugging Face models on AWS in the "Hugging Face on AWS" docs.

AWS Deadline Cloud is a fully managed batch compute service (job queuing, scheduling, auto-scaling GPU worker fleets) and its samples repository, aws-deadline/deadline-cloud-samples, includes a growing set of ready-to-submit job bundles built on Hugging Face models and libraries: batch LLM inference with vLLM, LoRA/QLoRA fine-tuning with Transformers + PEFT + bitsandbytes, text-to-image batch generation, FLUX.2 Klein LoRA training, LLM benchmark evaluation, and ESMFold protein structure prediction.

Changes

• docs/sagemaker/source/index.md — two new sections following the existing per-service pattern (short description + quickstart link):
  • Run batch inference with AWS Deadline Cloud under "Deploy models on AWS"
  • Train with AWS Deadline Cloud under "Train models on AWS"
• docs/sagemaker/source/tutorials/deadline-cloud/deadline-cloud-quickstart.md — new quickstart modeled on the Bedrock/JumpStart quickstarts: why Deadline Cloud, prerequisites, batch LLM inference with the vllm_batch bundle, LoRA/QLoRA fine-tuning with the hf_finetune_lora bundle, clean-up. Commands and parameters are taken from the sample READMEs.
• docs/sagemaker/source/_toctree.yml and docs/sagemaker/source/tutorials/index.md — navigation entries.

Note: I work on the AWS Deadline Cloud team and maintain the samples this references. Happy to adjust wording, structure, or link style (the quickstart links use absolute URLs to match the existing sections; they will resolve once the page is published).

---

Note

Low Risk
Documentation-only changes with no runtime, API, or security impact.

Overview
Documents AWS Deadline Cloud as another way to run Hugging Face workloads on AWS, alongside SageMaker, Bedrock, Inference Endpoints, and EC2/ECS/EKS.

The landing page (index.md) gains bullet links plus two sections in the same style as existing services: batch inference under “Deploy models on AWS” and training (LoRA/QLoRA and multi-step pipelines) under “Train models on AWS”. A new Deadline Cloud Quickstart tutorial covers prerequisites, vllm_batch JSONL inference, hf_finetune_lora submission, PEFT loading, and cleanup, with pointers to deadline-cloud-samples job bundles. _toctree.yml and tutorials/index.md add navigation to that quickstart.

^{Reviewed by Cursor Bugbot for commit 46b1c5b. Bugbot is set up for automated code reviews on this repo. Configure here.}