fix: correct prompt_cache_retention literal type from "in-memory" to "in_memory"

[alvinttang]

Summary

Fixes #2883.

The prompt_cache_retention parameter type declaration used "in-memory" (hyphen) as a valid Literal value, but the OpenAI API expects "in_memory" (underscore). Passing "in-memory" results in a 400 error from the API.

Changes

• src/openai/types/chat/completion_create_params.py — Literal["in-memory", "24h"] → Literal["in_memory", "24h"]
• src/openai/types/responses/response_create_params.py — same fix
• src/openai/types/responses/response.py — same fix
• src/openai/types/responses/responses_client_event_param.py — same fix
• src/openai/types/responses/responses_client_event.py — same fix
• src/openai/resources/responses/responses.py — all overloads updated
• src/openai/resources/chat/completions/completions.py — all overloads updated
• tests/api_resources/test_responses.py — test values updated to match
• tests/api_resources/chat/test_completions.py — test values updated to match

Reproduction

from openai import OpenAI
client = OpenAI()
# Before fix: raises 400 because "in-memory" is rejected by the API
response = client.responses.create(
    model="gpt-4o",
    input="hello",
    prompt_cache_key="my-key",
    prompt_cache_retention="in-memory",  # wrong — API returns 400
)
# After fix: use "in_memory" (underscore)
response = client.responses.create(
    model="gpt-4o",
    input="hello",
    prompt_cache_key="my-key",
    prompt_cache_retention="in_memory",  # correct
)

🤖 Generated with Claude Code