Introduce method that returns raw logits for individual estimators

[bejaeger]

Motivation and Context

Provide a way to write out logits of individual estimators to perform analyses on the logit level afterwards.

---

Public API Changes

• No Public API changes
• Yes, Public API changes (Details below)

---

How Has This Been Tested?

---

Checklist

• The changes have been tested locally.
• Documentation has been updated (if the public API or usage changes).
• A entry has been added to CHANGELOG.md (if relevant for users).
• The code follows the project's style guidelines.
• I have considered the impact of these changes on the public API.

---

[Copilot]

Pull Request Overview

This PR introduces a new method predict_raw_logits that returns raw logits for individual estimators without averaging or temperature scaling, enabling logit-level analysis of ensemble components.

Key changes:

• Added predict_raw_logits method to expose raw logits from individual estimators
• Extended _raw_predict and forward methods with return_raw_logits parameter to support the new functionality
• Added comprehensive test coverage for the new method

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/tabpfn/classifier.py	Implements predict_raw_logits method and extends internal prediction pipeline to support raw logits output
tests/test_classifier_interface.py	Adds test cases to validate the new predict_raw_logits method functionality

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[gemini-code-assist]

Code Review

This pull request introduces a useful feature to extract raw logits from individual estimators, which is valuable for deeper analysis. The changes are well-implemented, adding a new predict_raw_logits method and correctly propagating the necessary flags through the internal methods. The inclusion of a dedicated test case is also great.

My review includes a few suggestions to improve the clarity of the docstrings for the new boolean flags to avoid ambiguity, and a minor refactoring in the forward method to streamline the logic for building the post-processing pipeline. Overall, this is a solid contribution.

[carterprince]

hi -- are there plans to add this to regressor.py as well?