refactor(terraform): revamp GenAI & ML infrastructure to lightweight application-level IaC

[iht]

Description

This PR refactors and modernizes the Terraform infrastructure and pipeline scripts for the GenAI & Machine Learning inference solution guide (terraform/ml_ai and pipelines/ml_ai_python), following the pattern established for the ETL/Integration use case in #221.

Key Changes

• Lightweight Application Infrastructure (terraform/ml_ai):
  • Removed foundation-level resource creation (google_cloud_project, vpc_network, firewall_rules, regional_nat).
  • Simplified inputs to application variables: project_id, region, subnetwork, bucket_name, service_account_name, create_bucket, destroy_all_resources.
  • Upgraded Google Cloud Foundation Fabric module sources to v57.0.0.
  • Added modern Cloud Build service account ([PROJECT_NUMBER]-compute@developer.gserviceaccount.com) to Artifact Registry admin permissions.
• Pipeline Scripts & Runtime Compatibility (pipelines/ml_ai_python):
  • Generated dynamic 00_set_variables.sh environment configuration.
  • Aligned Beam SDK version in Dockerfile (apache/beam_python3.11_sdk:2.75.0) to match requirements.txt.
  • Added --no_use_public_ips, --number_of_worker_harness_threads=1, and flexible GPU accelerator configuration to 02_run_dataflow.sh.
  • Added Python 3.11 version matching documentation to pipelines/ml_ai_python/README.md.
• Documentation:
  • Updated terraform/ml_ai/README.md, pipelines/ml_ai_python/README.md, use_cases/GenAI_ML.md, and terraform/TERRAFORM_REVAMP_GUIDE.md.

Verification

• Deployed infrastructure, built container image with Gemma 2B model weights, and launched streaming Dataflow GPU job on Google Cloud.
• Verified prompt submission to Pub/Sub topic messages and received valid inference output from predictions-sub.
• Clean teardown executed via terraform destroy.