Deploy tabular foundation models to SageMaker endpoints

[shchur]

Issue #, if available:

Description of changes:

• implement TabularFoundationModel.deploy using the shared foundation-model deployment path
• add TabularEndpoint with predict and predict_proba APIs that send both prediction data and labeled train_data
• add a tabular foundation-model serving handler that caches Hugging Face weights at startup, fits a request-scoped TabularPredictor, and renders predictions/probabilities
• add serializer support, CPU deployment defaults, public exports, API docs, and unit coverage

Validation:

• 50 passed in the focused unit suite
• Ruff and git diff --check pass
• deployed mitra-classifier with AutoGluon 1.5.0 on ml.m5.4xlarge, invoked predictions and probabilities successfully, and verified endpoint/config/model cleanup

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.