fix(docs): link required sites for models without a coefficients file

[marcbal77]

What

Closes #200. HurdleInflammAge showed N/A in the Coefficients column of the model table because it has no model.file. Its required CpGs already ship in biolearn/data/Hurdle_CpGs.csv, so the docs now link to that file instead.

How

• New optional sites_file key in a model definition's model block, for models that have no coefficients file but do have a fixed list of required sites.
• HurdleAPIModel reads its CpG list from sites_file rather than a hardcoded filename, so the definition is the single source for both the model and the docs. The default is unchanged (Hurdle_CpGs.csv), so direct construction behaves as before.
• doc/generate.py links required sites when there is a sites_file and no file. Everything else in the generator is untouched.

Instantiating the model in the doc build (the approach floated in the issue) was avoided: HurdleAPIModel requires HURDLE_API_KEY to construct, which the docs build shouldn't depend on. Any future API-backed model can set sites_file and get the same treatment.

Output

Only one row of generated/model_table.csv changes:

- HurdleInflammAge,...,N/A
+ HurdleInflammAge,...,`required sites <https://github.com/bio-learn/biolearn/blob/master/biolearn/data/Hurdle_CpGs.csv>`_

model_usage.csv is identical.

Tests

• test_sites_loaded_from_definition_sites_file: sites come from the definition's file.
• test_missing_sites_file_falls_back_to_all_sites: the parameter is actually used, and an unknown file keeps the existing warn-and-fallback behavior.
• test_sites_file_exists: every sites_file in model_definitions exists in biolearn/data, so the docs link can't point at a missing file.

The two Hurdle tests fail against master and pass here. make check-format clean. make test: 205 passed, 5 skipped, 0 failed.

Depends on #218

Stacked on #218, which fixes the CI failure every PR currently hits (geo2r now returns a CAPTCHA page). Merge #218 first and this diff reduces to the commit above.

Note

doc/generate.py also crashes on master in generate_data_csv_from_yaml because the NHANES and FHS entries from #203 have no samples field. That is fixed separately in #219.