[Task] tabular-classification model support

Overview

Tabular classification models predict a categorical label from structured (tabular) data — rows of numeric and categorical features. These models are heavily used in enterprise and data analytics workflows. Unlike vision/NLP models, they typically use gradient-boosted tree or MLP architectures (e.g. XGBoost-backed ONNX exports, TabNet, FT-Transformer).

Target: top 2–8 models by HuggingFace downloads (>2k) covering representative architectures and business domains.

Agent Scenarios

• Risk / fraud detection agent: classify financial transactions as fraudulent or legitimate using structured account and transaction features
• Customer churn prediction agent: predict likelihood of customer churn from CRM feature tables to trigger retention workflows
• Medical diagnosis agent: classify patient records against structured clinical features (lab values, vitals, demographics)
• Lead scoring agent: rank and classify sales leads from CRM attributes to prioritize outreach

ModelKit Integration

Models must pass the full wmk pipeline on all EPs:

wmk config → wmk build (ONNX export) → wmk perf → wmk eval

Acceptance Criteria

• Identify top 2–8 tabular-classification models (>2k downloads) and add to model list
• All selected models pass wmk perf on CPU EP
• All selected models pass wmk eval with tabular dataset

No source file or test is named. Start by locating the model list, identify 2–8 tabular-classification models with more than 2k Hugging Face downloads, and run the listed wmk config, build, perf, and eval commands. Done means the selected models are listed and pass CPU performance and tabular-dataset evaluation.