fix(serve/bedrock): robust S3 manifest lookup and improved error handling

[Mattral]

fix(serve/bedrock): robust S3 manifest lookup and improved error handling

Summary

Robust S3 manifest lookup and clearer S3 error handling for Bedrock Nova models.

Files Changed

• bedrock_model_builder.py
• test_bedrock_model_builder.py

Description of Changes

• Try multiple candidate manifest locations:
  • dirname/manifest.json
  • dirname/output/manifest.json
  • parent/manifest.json
• Use ClientError (botocore) where available to detect missing keys/buckets and continue searching.
• Provide clear ValueError when a checkpoint cannot be located.
• Search manifest contents for checkpoint-like keys (any key containing "checkpoint"), with fallback to common names:
  • checkpoint_s3_bucket
  • checkpoint_s3_uri
  • checkpoint_s3_path
• Add unit tests covering success and manifest-not-found cases.
• Preserve original authors’ comments and intent; changes focus on robustness and observability.

Impact

• Prevents opaque failures during Bedrock model import when checkpoint manifest locations vary.
• Improves debuggability with clearer error messages and deterministic path resolution.
• No behavioral changes for non-Nova models.

Testing

• Added test_bedrock_model_builder.py (two tests).
• Ran tests locally in the dev container — both tests passed.
• Recommendation: Run CI to execute the new unit tests.

Review Notes

• Review candidate S3 key choices and error-message wording.
• Verify alignment with Bedrock/Training job artifact layout expectations.

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.