[codex] Use uploaded cache endpoint for Runner packages

[k82cn]

Summary

• Preserve the object-cache endpoint returned by package upload when Runner builds package URLs.
• Normalize grpc+tls:// endpoints to grpcs:// for executor-manager downloads.
• Add Runner storage tests for returned pod endpoints and TLS endpoint normalization.

Root Cause

Runner uploaded packages through object-cache, but rebuilt the package URL from the configured cache service endpoint instead of the returned ObjectRef endpoint. With multiple object-cache pods behind one ClusterIP service, executor-manager could download from a different cache pod and fail to find the package.

Verification

• uv run -n pytest tests/test_runner.py -q
• python3 -m py_compile sdk/python/src/flamepy/runner/storage.py sdk/python/tests/test_runner.py
• git diff --check

[gemini-code-assist]

Code Review

This pull request introduces a helper method _package_url_from_ref in CacheStorage to dynamically resolve and normalize the package URL from the uploaded object's reference, specifically handling the translation of grpc+tls:// to grpcs://. It also adds corresponding unit tests. The feedback suggests a cleaner and safer way to replace the protocol prefix using str.replace instead of manual string slicing.

[gemini-code-assist]

Using manual string slicing with a hardcoded length calculation like endpoint[len('grpc+tls://'):] is more error-prone and less readable than using str.replace with a limit of 1. If the prefix string is ever updated in the future, having to update both the startswith check and the slice length calculation can easily lead to bugs.

Using endpoint.replace("grpc+tls://", "grpcs://", 1) is cleaner and safer.

Suggested change

	if endpoint.startswith("grpc+tls://"):
	endpoint = f"grpcs://{endpoint[len('grpc+tls://'):]}"
	if endpoint.startswith("grpc+tls://"):
	endpoint = endpoint.replace("grpc+tls://", "grpcs://", 1)