fix(cli): show_reasoning works independently of streaming setting (#34209)

[Bartok9]

Closes #34209

Problem

In CLI mode, show_reasoning=true only displayed reasoning when streaming=true. Users who prefer non-streaming output (streaming: false) couldn't see reasoning content even with show_reasoning: true explicitly set.

# cli.py::_current_reasoning_callback (before)
if self.show_reasoning and self.streaming_enabled:
    return self._stream_reasoning_delta
# show_reasoning=True, streaming=False fell through to None

The agent had no reasoning sink and reasoning tokens were silently dropped.

Fix

The CLI already has a buffered-preview render path (_on_reasoning + _flush_reasoning_preview) used today for verbose=True, show_reasoning=False — emits compact [thinking] blocks on natural boundaries.

Reusing it for the non-streaming show_reasoning case restores expected behavior without changing the user's CLI output preferences.

Coverage matrix after fix

show_reasoning	streaming	verbose	reasoning render
True	True	any	live token-by-token box
True	False	any	buffered [thinking] preview ← #34209
False	any	True	minimal [thinking] preview
False	any	False	no reasoning render

Tests

• Updated test_show_reasoning_non_streaming_uses_final_box_only → ...uses_buffered_preview. The old assertion enshrined the bug (assertIsNone); the new one locks in the fix.
• 3 new tests: test_no_callback_when_all_options_off, test_no_callback_when_streaming_only, test_show_reasoning_takes_precedence_over_verbose.

$ python -m pytest tests/cli/test_reasoning_command.py
=== 58 passed in 1.11s ===

🎻 Co-authored-by: Cursor cursoragent@cursor.com

[alt-glitch]

Competing with #34224 for the same fix (#34209). Both route non-streaming reasoning through the buffered preview callback. This PR (Cursor co-authored) adds more test coverage.

[Bartok9]

Closing as duplicate of #34224 — thanks @alt-glitch and the reviewer pipeline for flagging the dup pattern (same as #34376 \u2192 #34241 earlier tonight).

@LeonSGP43's PR #34224 (opened ~54 min before mine) makes the same routing fix to _current_reasoning_callback AND \u2014 importantly \u2014 also marks _reasoning_shown_this_turn = True in the preview callback path, which prevents the final-response box from rendering a duplicate reasoning block. That's a real edge case my PR missed.

#34224 is the canonical fix; closing this one.

\u2014 Bartok9