fix(types): use in_memory for prompt_cache_retention

[MukundaKatta]

Summary

• update prompt_cache_retention literals from in-memory to in_memory across the Python SDK type surfaces and resource method signatures
• update responses and chat completions request tests to use the underscore form accepted by the API
• keep the exposed SDK surface consistent so typed usage no longer suggests a value the API rejects

Fixes #2883

Validation

• scanned src and tests to confirm there are no remaining prompt_cache_retention literals using in-memory
• compiled the touched source and test files successfully with PYTHONPYCACHEPREFIX=.pycache and PYTHONPATH=src
• attempted the targeted pytest files, but the local environment does not have the repo mock-server and async-extra setup needed for those suites to be reliable here

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bbdee6d0a5

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Keep streamed output text non-null before delta appends

Changing ResponseOutputText.text to optional allows response.content_part.added snapshots to carry text=None, but streaming accumulation still assumes a string and does content.text += event.delta in src/openai/lib/streaming/responses/_responses.py (lines 349-354). In streams where a content part starts as null and then receives response.output_text.delta events, this will raise a TypeError and terminate parsing; please preserve a non-null invariant for streamed output_text content (or normalize null to "" before appending).

Useful? React with 👍 / 👎.

[MukundaKatta]

Addressed @chatgpt-codex-connector review: normalize content.text to "" before += event.delta so response.content_part.added snapshots carrying text=None don't raise TypeError on the first delta. Streaming invariant (text is always a string after accumulation starts) is preserved.

[chatgpt-codex-connector]

Note

To use Codex here, create a Codex account and connect to github.