Fix/mlp embedding

[jhnwu3]

This pull request refactors the MLP model in pyhealth/models/mlp.py to simplify and centralize the embedding logic by introducing the new EmbeddingModel class. The changes improve maintainability and make the handling of different input types more consistent and modular.

Model architecture refactoring

• Replaced custom embedding and linear layer management in the MLP model with a dedicated EmbeddingModel class for handling all embedding logic, reducing code duplication and complexity. [1] [2]
• Updated the forward method to preprocess inputs, track reshaping information, and delegate embedding operations to EmbeddingModel, followed by pooling and concatenation of patient embeddings. [1] [2] [3]

Test data update

• Modified the test sample in tests/core/test_mlp.py to add an additional condition to the input data, ensuring coverage of the updated embedding handling.