test: add unit tests for RNN and MultimodalRNN models

[haoyu-haoyu]

Summary

Add tests/core/test_rnn.py with 12 test cases covering the RNN and MultimodalRNN models, which previously had no unit tests.

TestRNN (8 tests)

Test	What it verifies
test_model_initialization	Correct attributes (embedding_dim, hidden_dim, feature_keys, label_key)
test_model_forward	Output dict contains loss, y_prob, y_true, logit with correct shapes
test_model_backward	Gradients propagate to at least one parameter
test_model_with_embedding	embed=True returns patient embeddings of expected dimension
test_custom_hyperparameters	Non-default embedding_dim/hidden_dim work correctly
test_rnn_type_lstm	LSTM cell type variant produces valid output
test_rnn_type_vanilla	Vanilla RNN cell type variant produces valid output
test_bidirectional	Bidirectional RNN with down-projection produces valid output

TestMultimodalRNN (4 tests)

Test	What it verifies
test_model_initialization	Sequential vs non-sequential feature classification is correct
test_model_forward	Mixed modalities (sequence + multi_hot + tensor) forward pass works
test_model_backward	Gradient propagation through mixed-modality architecture
test_model_with_embedding	Embedding dimension = (n_seq × hidden_dim) + (n_nonseq × embedding_dim)

Pattern

Follows the established test pattern from test_mlp.py and test_tcn.py:

• Synthetic data via create_sample_dataset()
• Standard unittest.TestCase structure
• Tests initialization, forward, backward, embedding extraction, and hyperparameter variants

Ref #425