feat: expose configurable websocket max_size limit

[anneheartrecord]

Summary

Both websocket transports hard-coded max_size=None, which disables the websockets library's incoming message-size limit and gives applications no way to bound it without forking or monkeypatching the SDK:

• OpenAIResponsesWebSocketOptions only exposed ping_interval / ping_timeout; _open_websocket_connection passed max_size=None.
• TransportConfig (Realtime) only exposed ping_interval / ping_timeout / handshake_timeout; _create_websocket_connection passed max_size=None.

This adds an optional max_size: int | None to both option types and passes it through to websockets.connect when set, mirroring the existing keepalive knobs. Long-lived agent processes running behind proxies or in memory-constrained containers can now bound per-message memory, e.g.:

responses_websocket_options={"max_size": 8 * 1024 * 1024}
OpenAIRealtimeWebSocketModel(transport_config={"max_size": 8 * 1024 * 1024})

The default behavior is unchanged — when max_size is not provided the connection is still opened with no limit, so this is purely opt-in.

Test plan

• Added test_websocket_model_passes_max_size_to_connect (Responses) and test_max_size_config_is_applied (Realtime) verifying the value reaches websockets.connect.
• The existing test_websocket_model_open_websocket_connection_disables_message_size_limit (asserting the default stays None) still passes.
• make lint, make format, mypy and pyright on the changed modules, and the full tests/models/test_openai_responses.py + tests/realtime/test_openai_realtime.py suites (181 passed) are green. Docs build succeeds.

Issue number

Closes #3644

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've run make lint and make format
• I've made sure tests pass