sktime - Interfacing Foundation models

sktime provides a unified Python API for time series forecasting, but its support for modern foundation models, large pretrained models capable of zero-shot forecasting, is still maturing. Several high-value models, such as Google TimesFM and IBM TinyTimeMixer, are not yet interfaced in sktime; existing foundation model adapters lack probabilistic output (predict_interval, predict_quantiles), and there is no shared wrapper to reduce the boilerplate cost of adding new models from the Hugging Face Hub. This project directly addresses those gaps. The primary deliverable is one production-ready foundation model adapter fully integrated into sktime's BaseForecaster API, with fit, predict, predict_interval, and predict_quantiles support, CI-passing tests, soft-dependency handling, and a usage tutorial.