fix(realtime): preserve Unicode characters in realtime calls session serialization

[piyushrajyadav]

Summary

This PR ensures non-ASCII Unicode characters (such as Cyrillic, Chinese, Japanese, Arabic, etc.) in the session payload for realtime.calls.create (sync & async) are not converted to \uXXXX escape sequences during serialization.

Problem

When passing instructions, names, or other parameters containing non-ASCII characters to client.realtime.calls.create(..., session=...), Python's json.dumps() defaults to ensure_ascii=True, converting characters like Привет into \u041f\u0440\u0438\u0432\u0435\u0442. This significantly bloats the payload size and leads to excess token consumption upstream.

Solution

• Added ensure_ascii=False to json.dumps() in both _Calls.create and _AsyncCalls.create in src/openai/lib/_realtime.py.
• Added unit tests in tests/test_realtime_calls_unicode.py to verify that Unicode characters remain unescaped UTF-8 in both sync and async request payloads.

Fixes #2428

How it was tested

Ran the test suite locally:

python -m pytest -s -vv -n0 tests/test_realtime_calls_unicode.py

Output:

============================= test session starts =============================
platform win32 -- Python 3.14.7, pytest-9.1.1, pluggy-1.6.0
rootdir: E:\open source\open ai\openai-python
configfile: pyproject.toml
plugins: anyio-4.14.2, inline-snapshot-0.35.4, langsmith-0.10.17, asyncio-1.4.0, xdist-3.8.0
asyncio: mode=Mode.AUTO, debug=False, asyncio_default_fixture_loop_scope=session, asyncio_default_test_loop_scope=function
collecting ... collected 2 items

tests/test_realtime_calls_unicode.py::TestRealtimeCallsUnicode::test_realtime_create_unicode_session_not_escaped PASSED
tests/test_realtime_calls_unicode.py::TestRealtimeCallsUnicode::test_async_realtime_create_unicode_session_not_escaped PASSED

============================== 2 passed in 1.91s ==============================

Linting:

python -m ruff check src/openai/lib/_realtime.py tests/test_realtime_calls_unicode.py
All checks passed!