fix(lib): guard parse_response against response.output=None

[nicha16]

Fixes #3459

Problem

parse_response in src/openai/lib/_parsing/_responses.py iterates response.output without guarding against None:

for output in response.output:   # TypeError if response.output is None

When a completed Response has output=None, this raises TypeError: 'NoneType' object is not iterable. Because parse_response is invoked from the streaming accumulator (lib/streaming/responses/_responses.py::accumulate_event → parse_response), the exception crashes the entire stream rather than surfacing as a handleable result — and it originates deep in the SDK, so it isn't catchable as a normal API error.

output can be None for incomplete / failed / reasoning-only responses, and via some proxy/aggregator backends.

Fix

Guard the iteration:

-    for output in response.output:
+    for output in (response.output or []):

A None/absent output now yields a ParsedResponse with an empty output list instead of crashing.

Test

Adds tests/lib/responses/test_parse_response_none_output.py, a focused regression test that calls parse_response with a Response whose output is None.

Validated in isolation:

• Without the guard: the test fails with exactly TypeError: 'NoneType' object is not iterable at _parsing/_responses.py (reproduces the bug).
• With the guard: the test passes.

The change is one line plus a test; happy to adjust naming/placement to match your conventions.