fix: predict output quality — stray y, index-keyed intervals, horizon cap (#536)

[Shashankss1205]

Fixes #536. Three predict output-quality issues:

NB-18 — y forwarded to non-annotators (raw TypeError)

predict(..., y_dataset=...) on a forecaster raised BaseForecaster.predict() got an unexpected keyword argument 'y'. y at predict is only for annotators/detectors; it's now passed only when predict's signature accepts it, and otherwise dropped with a warnings entry explaining why.

NB-21 — interval/variance modes dropped the time index

predict_interval / predict_var serialized the DataFrame with orient="list", so values came back as bare arrays with no time labels — while predict keys by period. Switched to orient="index", so every mode is index-keyed:

"intervals": {"1961-01": {"...0.8_lower": 357.3, "...0.8_upper": 476.7}, ...}

NB-22 — unbounded horizon flooded the response

predict(horizon=1000) returned 1000 inline rows. Inline forecast rows are capped at 500 with a predictions_truncated: {shown, total, note} marker; normal horizons are never affected.

Testing

• New tests/test_predict_output.py (5 tests): stray y dropped with a warning (not a crash); predict_interval and predict_var are index-keyed dicts; horizon=1000 capped with a marker; horizon=12 untouched.
• Full suite: 280 passed.

🤖 Generated with Claude Code

[Shashankss1205]

Closing: this PR's content is already on main as commit 4d870f0 (identical patch-id; it was landed directly rather than through the PR merge button). Nothing further to merge.