fix: extend output_text to include code_interpreter and shell output

[roli-lpci]

Summary

The output_text convenience property only aggregates text from message output items. When a response contains code_interpreter_call or shell_call_output items with textual results but no accompanying message, output_text returns an empty string.

This extends the property to also collect:

• code_interpreter_call: log output from OutputLogs.logs
• shell_call_output: stdout from Output.stdout

Root Cause

Despite the report in #2870, the SDK code between v2.19.0 and v2.20.0 is unchanged for this codepath — output_text has always only traversed message-type output items:

# Before (v2.19.0 through v2.24.0 — identical)
for output in self.output:
    if output.type == "message":  # ← only messages
        for content in output.content:
            if content.type == "output_text":
                texts.append(content.text)

Code interpreter results live in code_interpreter_call items via outputs[].logs, which is a separate data path that output_text never traversed. This is a feature gap in the convenience property, not a deserialization regression.

Changes

• src/openai/types/responses/response.py — Extend output_text to also iterate code_interpreter_call log outputs and shell_call_output stdout. Updated docstring to reflect the expanded aggregation.
• tests/lib/responses/test_responses.py — 6 new tests:
  • Mixed message + code interpreter output (ordering preserved)
  • Code interpreter only (no message item)
  • outputs: null (graceful empty string)
  • Image-only outputs (correctly skipped)
  • Shell call stdout

Scope note

This PR covers code_interpreter_call and shell_call_output as the two tool types with unambiguous textual output. Other tool types (mcp_call, apply_patch_call_output) could be added in a follow-up if desired.

Testing

All 110 tests/lib/ tests pass, including the 6 new ones.

Fixes #2870