fix: reject h < 1 in ConformalIntervals at construction

[cjck944084735-dot]

Summary

ConformalIntervals validated n_windows but not h, so ConformalIntervals(h=0) constructed successfully and only failed later — at interval-computation time — with a bare ZeroDivisionError from (n_samples - 1) // h in _conformity_scores, pointing nowhere near the bad argument.

Fixes #1221.

The fix

ConformalIntervals.__init__ now rejects h < 1 with a clear message, mirroring the existing n_windows < 2 check:

>>> ConformalIntervals(h=0)
ValueError: h must be at least one to compute conformal intervals

Testing

• New test_conformal_intervals_invalid_h in tests/test_models.py: asserts h=0 and a negative h raise ValueError at construction.
• Verified the full flow still works for a valid h (Naive + ConformalIntervals(h=4) produces lo-80/hi-80 as before), and that the bare ZeroDivisionError from the issue is gone.
• Ran the conformal test subset of tests/test_models.py: the only failures are 3 pre-existing ones (test_arima_conformal_prediction, test_mfles_conformal_prediction, test_sklearn_model_with_conformal_intervals) that fail identically on unpatched main in this environment (numba-version related), not introduced by this change.
• ruff check (repo config: F) passes on both changed files.

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[cjck944084735-dot]

Transparency note: this fix was drafted with the help of an AI coding agent under the direction of the account owner, who reproduced the issue, reviewed the diff line by line, ran the verification (repro before/after, the new regression test, and the conformal test subset), and takes responsibility for the change. Happy to answer any questions about the details.