perf: defer eager type/streaming imports at import openai

[harsheet-shah]

Summary

import openai eagerly imports the entire Assistants beta type tree and the streaming helpers, building hundreds of pydantic models that most callers never use. This dominates cold-start / first-import latency in serverless and hosted-agent environments (e.g. containers that boot, import the SDK, then serve a single Responses/Chat request).

This PR makes the top-level openai package expose the rarely-needed surfaces lazily while keeping the public API and static typing identical:

• openai.types is now a LazyProxy (mirrors the existing openai.resources proxy) instead of from . import types.
• pydantic_function_tool, AssistantEventHandler, AsyncAssistantEventHandler, and the ReconnectingEvent / ReconnectingOverrides websocket types resolve through a module-level __getattr__ (PEP 562).
• openai.lib defers _tools / _parsing the same way.
• All names remain statically importable via TYPE_CHECKING re-exports, so IDEs / pyright are unaffected, and from openai import * is unchanged.

Azure / Bedrock module-client support and every existing public object are untouched.

Measurements

openai 3.11.0, warm best-of-3, import openai on top of an already-imported pydantic + httpx2:

	import openai	eager openai.* submodules
main (eager)	~1,944 ms	835
this PR (lazy)	~1,101 ms	497

~43% faster first import; the Assistants beta subtree (309 modules) is no longer built at import time. It loads transparently on first access (openai.beta, openai.types.beta, etc.).

Tests

Adds tests/test_lazy_imports.py:

• import openai no longer eagerly loads openai.types.beta or openai.lib.streaming.
• The lazy top-level exports resolve to the same objects as their canonical modules.
• The types proxy loads the real module on first attribute access.

Existing tests/test_module_client.py and tests/lib/test_old_api.py continue to pass, and a full import smoke over every openai.types.* subtree + from openai import * verifies no exported name regressed.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bd68d7aad5

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[eavanvalkenburg]

This would definitely improve import performance across the board, and complements a fix I made in our package: microsoft/agent-framework#8219