fix: read workload token as utf-8

[Ghraven]

Summary

• read the Kubernetes workload identity subject token file with explicit UTF-8 encoding
• update the matching test fixture write to use the same encoding

Before / after

Before, k8s_service_account_token_provider() used the platform default text encoding when reading the mounted token file.
After, it reads the token as UTF-8 explicitly, matching the text-file behavior expected by the test.

Verification

• python -m py_compile src/openai/auth/_workload.py tests/test_auth.py
• git diff --check