make MLP compatible with the new pyhealth 2.0 loaders

[jhnwu3]

This pull request refactors the MLP model in pyhealth/models/mlp.py to support the new SampleDataset format and simplifies the model's interface and feature handling. It also adds support for a new processor type in the embedding model and improves code clarity and maintainability throughout. The most important changes are grouped below:

Dataset and Model Interface Modernization:

• Refactored the MLP model to use SampleDataset instead of SampleEHRDataset, updating constructor arguments and usage examples for a simpler and more unified interface. Feature and label keys are now inferred from the dataset, removing the need to manually specify them. [1] [2] [3] [4] [5]

Feature Handling and Forward Logic Simplification:

• Rewrote the forward method in MLP to dynamically handle different feature types and tensor shapes, including automatic creation of embedding or linear layers as needed, and streamlined pooling logic for both categorical and numerical inputs.

Processor and Embedding Model Extension:

• Added support for TensorProcessor in the embedding model, enabling flexible handling of tensor-based features by automatically determining input size from sample data. [1] [2]

Documentation and Code Clarity:

• Improved and clarified docstrings and comments in the MLP model, including more detailed explanations of supported input formats and pooling methods. [1] [2]

Miscellaneous Improvements:

• Minor formatting and registry logic improvements in the processor module for future extensibility.