refactor(core): promote ModelOutputThunk.thinking and deprecate _thin…

[exequeryphil]

Pull Request

Issue

Fixes #1217

Description

• Promoted ModelOutputThunk.thinking to the public reasoning-trace field in mellea/core/base.py, with a temporary deprecated _thinking alias that emits DeprecationWarning.
• Updated reasoning capture across backends (openai, litellm, ollama, watsonx) to write into mot.thinking.
• Updated safety/guardian requirement logic to read mot.thinking.
• Updated docs to reference result.thinking instead of result._thinking.
• Updated backend unit tests to assert mot.thinking, and added core tests for the new public field plus deprecated alias behavior.
• Resolved merge conflict in test/core/test_base.py by keeping both your new thinking-related tests and upstream base tests.

Testing

• [X ] Tests added to the respective file if code was changed
• [X ] New code has 100% coverage if code was added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Attribution

• AI coding assistants used

---

Adding a new component, requirement, sampling strategy, or tool?

If your PR adds or modifies one of the types below, check the matching box. A checklist of type-specific review items will be posted as a comment.

• Component
• Requirement
• Sampling Strategy
• Tool

NOTE: Please ensure you have an issue that has been acknowledged by a core contributor and routed you to open a pull request against this repository. Otherwise, please open an issue before continuing with this pull request.

[akihikokuroda]

Looks good. A minor issue is reported.

Missing: Test for deprecated setter with warning

The tests verify reading _thinking generates a deprecation warning, but there's no test verifying the setter also warns. The implementation includes both getter and setter (which is correct), but the test only covers the getter path:

Current tests cover:
✅ Reading: mot._thinking (should warn)
❌ Not tested: mot._thinking = "value" (should warn)

Recommendation: Add one small test (no need for separate test, can extend existing):
def test_mot__thinking_deprecated_alias_warns_on_read_and_write():
mot = ModelOutputThunk(value="x")
with pytest.warns(DeprecationWarning):
mot._thinking = "reasoning trace" # Setter should also warn
with pytest.warns(DeprecationWarning):
value = mot._thinking # Getter warns
assert value == "reasoning trace"

[akihikokuroda]

@exequeryphil Please run:

uv run ruff format
uv run ruff check --fix

Thanks!

[ajbozarth]

This is looking solid, but I'm actually circling back on a design decision before we merge this. @planetf1 @jakelorocco I'm no longer sure it's worth including a deprecation shim for a private var like _thinking. I haven't be including deprecation shims for private vars in #1191 or my planned implementation for #1215 that I'm in the middle of.

If we do decide to keep the deprecation shim there are a few nits I'd raise, which I'll include a comments below

[ajbozarth]

Suggested change

	def test_mot__thinking_deprecated_alias_warns_on_read():
	def test_mot_thinking_deprecated_alias_warns_on_read():

no need to double underscore here, it just makes it less readable

[ajbozarth]

Suggested change

	def test_mot__thinking_deprecated_alias_write_sets_public_field():
	def test_mot_thinking_deprecated_alias_write_sets_public_field():

same here

[ajbozarth]

IIRC I don't think this is the correct docstring format for the project (Google style)

[ajbozarth]

This should have the same warn as read