fix(types): use in_memory for prompt cache retention - #3024

Fixes #2883.

Summary

• switch every generated prompt_cache_retention literal from "in-memory" to "in_memory" across chat completions, responses, and response event types
• update the generated API resource tests so the typed request examples now use the value the API actually accepts
• add a focused regression that resolves type hints for the request params, response models, response event models, and public resource methods to lock in the underscore form

Testing

• .venv\\Scripts\\python -m pytest tests/test_prompt_cache_retention.py tests/api_resources/test_responses.py::test_parse_method_in_sync -q -n 0
• .venv\\Scripts\\python -m ruff check tests/test_prompt_cache_retention.py src/openai/resources/chat/completions/completions.py src/openai/resources/responses/responses.py src/openai/types/chat/completion_create_params.py src/openai/types/responses/response.py src/openai/types/responses/response_create_params.py src/openai/types/responses/responses_client_event.py src/openai/types/responses/responses_client_event_param.py tests/api_resources/chat/test_completions.py tests/api_resources/test_responses.py

Notes

• The broader generated API resource suites still expect the project mock server on 127.0.0.1:4010, and some async variants also require the optional aiohttp extra, so this local validation stayed on the touched no-network regression plus the existing signature-sync check.