Metric for long-term forecast

For long-term forecast for physical systems which have stochatsitc components (e.g., chaotic systems) or turbulences, MSE-based metrics on long-term trajectory are not meaningful. We should consider bringing metrics to capture power spectrum of the solution in long-term forecast, e.g. as 'climate forecast' in [CTF](https://arxiv.org/pdf/2510.23166)