fix(types): use in_memory prompt cache retention literal

[MukundaKatta]

Summary

• replace the incorrect prompt_cache_retention literal value in generated Python types
• update request/response model surfaces to use in_memory instead of in-memory
• refresh request expectation tests for chat completions and responses

Testing

• Unable to run locally in this checkout because pytest_asyncio is not installed

Closes #2883

[MukundaKatta]

Superseded by #3108 which covers the same in_memory prompt_cache_retention literal fix with additional test + SDK-surface updates (+170/-48 vs this +41/-41). Closing the narrower copy.