[Model Integration] Add Automatic Piecewise Linear Regression (APLR) to TabArena

### Proposer
* **Algorithm Name:** Automatic Piecewise Linear Regression (APLR)
* **PyPI Package:** https://pypi.org/project/aplr/
* **Repository:** https://github.com/ottenbreit-data-science/aplr

### Context & Motivation
Following the contribution guidelines, I am opening this issue to discuss integrating APLR into TabArena's tabular benchmark suite.

APLR is a parametric, component-wise boosting-based machine learning framework designed for tabular datasets. Instead of relying on decision trees, it constructs models using linear or piecewise linear terms. This formulation yields continuous, smooth prediction functions compared with the step-functions typical of tree-based additive architectures (such as gradient boosters or EBMs). This structural behavior prevents discontinuous jumps in feature effects, making it a relevant candidate for TabArena's suite of interpretable baselines.

Shapes of main and two-way interaction effects can be extracted and plotted. Additionally, an APLR model can be explicitly represented as a regression formula, and the framework has the capability to model higher-depth interactions if required.

### Technical Characteristics
* **Smooth Formulations:** Employs linear or piecewise linear terms to guarantee continuous prediction surfaces across feature spaces.
* **Model Transparency:** Allows extraction of feature importance, main effects, and interactions directly from the parametric terms.
* **Task Coverage:** Native implementations optimized for tabular regression and binary classification tasks.
* **Backend:** Implemented with a C++ core paired with a lightweight Python wrapper interface.

### Proposed Integration Path
I plan to fork the repository and submit a pull request that covers:
1. Adding `aplr` to the project's package dependency configurations.
2. Creating the standardized model wrapper under `packages/tabarena/src/` to map `.fit()`, `.predict()`, and `.predict_proba()` functionality.
3. Defining a baseline hyperparameter search space configuration profile.

### Questions for Maintainers
Before diving into the implementation, I would appreciate your guidance on a few repository preferences:
1. Should `aplr` be integrated as an optional framework extra, or included directly within the baseline evaluation matrix?
2. Does the proposed integration path align with your architectural goals, and is there a specific abstract wrapper class you prefer I inherit from?

Thank you for your feedback, and I look forward to contributing APLR to the framework!