Add Empirical Gaussian Processes tutorial (#3331)

Summary:
Pull Request resolved: #3331

Adds a self-contained tutorial demonstrating the Empirical Gaussian Process (Lin et al., ICML 2026, arXiv:2602.12082) on two real 1D forecasting problems: the S&P 500 index and Mauna Loa atmospheric CO2. For each series it compares an Empirical GP (prior mean/covariance learned from historical curves) against a handcrafted GP baseline built from the integrated-white-noise kernel added in the parent diff.

- `tutorials/empirical_gaussian_processes/empirical_gaussian_processes.ipynb`: uses only public botorch APIs (`botorch.models.empirical_gps` and `botorch.models.kernels`), loads data locally, honors `SMOKE_TEST`, and renders a two-panel figure plus an RMSE/NLL comparison table. The handcrafted baselines use `IntegratedWhiteNoiseKernel(order=...)`: the financial model is a geometric-Brownian-motion model with `order=1`, and the climate trend is a sum of `order=1`, `order=2`, and `order=3` terms.
- `tutorials/data/sap500.csv` and `tutorials/data/co2_daily_mlo.csv`: the two datasets, redistributed under the CC0 license.
- Registers the tutorial under "Gaussian Process Models" in `website/tutorials.json`.

Reviewed By: SebastianAment

Differential Revision: D109472635

fbshipit-source-id: c422c83d7356c749eff6297a5210b90a9890dff0

Author: Jihao Andreas Lin