Add PTB-XL dataset support (ECG)

[AxelNoun]

Contributor: AxelNoun (GitHub: @AxelNoun) — external contributor, no NetID
Contribution Type: New Dataset + Task
Description: Adds PTB-XL (12-lead ECG, PhysioNet open access) as a
PyHealth dataset, with a 5-superclass multi-label diagnosis task and a
split helper using the official stratified folds. Motivated by the CPBench
effort: it gives conformal prediction a multi-label ECG benchmark that needs
no credentialing, with metadata suited to covariate/label-shift splits.
Based on Wagner et al., Scientific Data 2020
(https://www.nature.com/articles/s41597-020-0495-6).

Files to Review:

• pyhealth/datasets/ptbxl.py — dataset implementation
• pyhealth/datasets/configs/ptbxl.yaml — table config (template)
• pyhealth/tasks/ptbxl.py — 5-superclass multi-label task
• pyhealth/datasets/splitter.py — split_by_strat_fold helper
• tests/core/test_ptbxl.py — tests with synthetic fixtures
• docs/api/datasets/pyhealth.datasets.ptbxl.rst, docs/api/tasks/pyhealth.tasks.ptbxl.rst
• examples/ecg/ptbxl/ptbxl_superclass_quickstart.py

Implementation notes:

• wfdb is an optional extra (pip install pyhealth[ptbxl]), lazily imported.
  Happy to make it a hard dependency if you'd prefer.
• Derived metadata CSV goes to MODULE_CACHE_PATH, not root. Filenames include
  sampling rate and a short hash of the resolved data root so different roots
  never share a cache; reuse also validates signal_file ⊆ data_root.
• Resolved cache YAML (new vs EEGBCI): BaseDataset only accepts
  config_path, so PTB-XL writes a root+rate-keyed YAML next to the derived CSV
  (ptbxl-config-{rate}hz-{root_hash}.yaml) before super().__init__(), with
  file_path already pointing at the derived metadata CSV. EEGBCI instead
  patches self.config.tables[...].file_path after super(). Open to aligning
  with maintainers' preferred pattern.
• Unrelated pre-existing # noqa: E402 on the EEGBCI import line in
  pyhealth/datasets/__init__.py — left untouched (out of diff).
• Signals returned as (n_leads, n_samples), matching EEGBCI's convention.
• Age ≥90 is encoded as 300 per HIPAA; kept raw with age_is_censored,
  distinct from age_is_missing (written as pd.NA).
• Empty label sets (407/21799) dropped by default via drop_empty_labels.
• Fixture under test-resources/ptbxl/ is synthetic WFDB (.hea/.dat)
  plus tiny CSVs covering censored age, missing age, multilabel, and empty superclass.

Testing / CI:

• Local: pytest tests/core/test_ptbxl.py — 17 passed (with wfdb installed).
• Without wfdb / pyhealth[ptbxl], the 2 waveform I/O tests skip; metadata /
  task / e2e event-read tests still run.
• pixi test feature pins wfdb so make testall under pixi covers waveform I/O;
  CI paths that do not install that extra will not exercise load_ptbxl_record.

Follow-up: 71-SCP-code classification and age regression are structured
for but not implemented here — planned as a separate PR.

Test plan

• pytest tests/core/test_ptbxl.py — 17 passed (with wfdb)
• python tools/check_pr_rules.py --base upstream/master --head HEAD
• CI make testall / contribution-rules on the PR

[AxelNoun]

Note for reviewers: there is an unrelated pre-existing # noqa: E402 on the EEGBCI import line in pyhealth/datasets/__init__.py; left untouched because it is out of this PR's diff.