Add SenBench-AQ-NO2-S5P and SenBench-AQ-O3-S5P datasets from SentinelBench

[wangyi111]

This PR adds SenBench-AQ-NO2-S5P and SenBench-AQ-O3-S5P datasets from SentinelBench.

The SenBench-AQ-NO2-S5P and SenBench-AQ-O3-S5P datasets are level-3 datasets from the SentinelBench benchmark.
It contains Sentinel-5P NO2/O3 images and EEA NO2/O3 maps for the air pollutants regression task.
It supports both static (1 image / location, annual mean) and time series (~4 images / location, seasonal mean) mode, the former is used in the original benchmark.

Dataset features:
* task: dense regression
* # samples: 1480/493/494 (train/val/test, static mode)
* image resolution: 56x56 (GSD 1km)
* label resolution: 56x56 (GSD 1km)
* mode: annual (static) or seasonal (time series)
* modality: NO2 or O3

Dataset format:
* images: 1 band Sentinel-5P NO2/O3 images (GeoTIFF)
* labels: EEA NO2/O3 maps (GeoTIFF)