feat!: improve consistency of post-training API endpoints

[eoinfennessy]

Summary

Updates post-training job endpoints to include {job_uuid} in the URL path instead of using it as a query parameter, while improving consistency with other Llama Stack APIs, and general adherence to RESTful conventions.

Changes

• POST /post-training/job/cancel → POST /post-training/jobs/{job_uuid}/cancel
• GET /post-training/job/status → GET /post-training/jobs/{job_uuid}/status
• GET /post-training/job/artifacts → GET /post-training/jobs/{job_uuid}/artifacts

Motivation

• Makes API consistent with other Llama Stack APIs, e.g.
  • Batches API: /batches/{batch_id}/cancel
  • Eval API: /eval/benchmarks/{benchmark_id}/jobs/{job_id}
  • Files API: /files/{file_id}
  • Conversations API: /conversations/{conversation_id}
• Aligns with OpenAI API, which uses POST /fine_tuning/jobs/{fine_tuning_job_id}/cancel for their fine-tuning API
• Follows REST best practices:
  • Resource IDs in URL path (not query parameters)
  • Plural resource names (/jobs/ instead of /job/)
• Adheres to Google API Improvement Proposals (AIPs):
  • AIP-136: Custom methods (cancel pattern)
  • AIP-122: Resource names (plural nouns)
• Fixes issue introduced in feat: migrate post_training API to FastAPI router #4496, where URL paths were not updated in the FastAPI routes.

Breaking Change

This is a breaking change for the v1alpha API. Users will need to:

• Regenerate client SDKs
• Update direct HTTP API calls to use new endpoint paths

Testing

• Manually verified by running server and Swagger UI

[leseb]

wondering if we should adhere to OpenAI fine_tuning API in the future too.