Add missing citations in PyHealth code

[lehendo]

Added to TFMTokenizer, EHRMambaCEHR, CEHR embeddings, comprehensiveness/sufficiency metrics, MLE, and NNAAR. Good context for AI when debugging/needs a reference.

[Copilot]

Pull request overview

Adds literature/code citations and usage examples across several PyHealth models and evaluation metrics to improve traceability and provide better debugging/reference context in the generated docs.

Changes:

• Added paper/code citations to TFMTokenizer, EHRMambaCEHR, and CEHR embedding docstrings.
• Added citations for interpretability (ERASER) and generative metrics (TSTR/MLE and NNAAR).
• Updated Sphinx API docs and an example script header to point readers to the new references.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
pyhealth/models/tfm_tokenizer.py	Adds TFM-Tokenizer paper + code citation in the model docstring.
pyhealth/models/ehrmamba_cehr.py	Adds EHRMAMBA paper citation and an end-to-end usage example to the class docstring.
pyhealth/models/cehr_embeddings.py	Adds EHRMAMBA/Odyssey embedding citation and a small usage example.
pyhealth/metrics/interpretability/sufficiency.py	Adds ERASER paper citation to the sufficiency metric docstring.
pyhealth/metrics/interpretability/comprehensiveness.py	Adds ERASER paper citation to the comprehensiveness metric docstring.
pyhealth/metrics/generative/utility.py	Adds TSTR protocol citation to MLE utility metric docstring.
pyhealth/metrics/generative/privacy.py	Adds NNAAR paper citation to privacy metric docstring.
examples/conformal_eeg/test_tfm_tuev_inference.py	Adds a pointer to the TFMTokenizer citation for the inference example.
docs/api/metrics/pyhealth.metrics.generative.rst	Updates the generative metrics API docs intro to reference per-function docstrings.

Comments suppressed due to low confidence (2)

pyhealth/models/tfm_tokenizer.py:731

• Line contains trailing whitespace in the class docstring (blank line with spaces). This can trigger linting failures (e.g., W291) and makes diffs noisier.

    The model expects two inputs:
        - STFT spectrogram: shape (batch, n_freq, n_time)
        - Raw temporal signal: shape (batch, n_samples)
    
    Args:

pyhealth/models/ehrmamba_cehr.py:32

• The docstring's Args section is incomplete: init accepts additional parameters (type_vocab_size, max_num_visits, time_embeddings_size, visit_segment_vocab) that are not documented here, which can mislead users reading the API docs.

    Args:
        dataset: Fitted :class:`~pyhealth.datasets.SampleDataset` with MPF task schema.
        vocab_size: Concept embedding vocabulary size (typically ``task.vocab.vocab_size``).
        embedding_dim: Hidden size (``hidden_size`` in CEHR embeddings).
        num_layers: Number of :class:`~pyhealth.models.ehrmamba.MambaBlock` layers.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.