feat: Deployment - Add AWS support

[rinat-akhmetov]

Add support for deploying Tangle on AWS infrastructure.

Target architecture:

• ECS/Fargate for running pipeline container tasks
• ECS/Fargate for hosting the backend (API Server + Orchestrator)
• S3 for artifact and log storage
• Aurora Serverless for database

Key components needed:

• S3 storage provider (similar to HuggingFaceRepoStorageProvider)
• ECS/Fargate launcher (implementing ContainerTaskLauncher interface)
• AWS-specific entry point script
• IAM role configuration

Related: HuggingFace deployment (#32), Kubernetes launcher support

[Ark-kun]

Hello, Rinat.
Thank you for creating the issue and providing the list of solutions and plans. Great ideas.

S3 storage provider (similar to HuggingFaceRepoStorageProvider)

A closer provider is https://github.com/Cloud-Pipelines/sdk/blob/master/cloud_pipelines/orchestration/storage_providers/google_cloud_storage.py

ECS/Fargate launcher (implementing ContainerTaskLauncher interface)

We could also consider EKS. Similar to class KubernetesWithGcsFuseContainerLauncher(_KubernetesContainerLauncher):

tangle/cloud_pipelines_backend/launchers/kubernetes_launchers.py

Line 511 in 26456ad

class KubernetesWithGcsFuseContainerLauncher(_KubernetesContainerLauncher):

AWS-specific entry point script

I'll polish the GCP startup script soon and we can align AWS startup script with it.

HuggingFace deployment

HuggingFace code is a bit messy. But it's currently available here:
https://huggingface.co/spaces/TangleML/tangle/tree/main/huggingface_overlay