pydata/xarray ? reverse-engineered prompt

Build me a Python library for working with multi dimensional scientific data that feels easy like pandas, but for arrays with names, coordinates, and metadata.

I want users to be able to create array and dataset objects, label dimensions like time or latitude, select data by those labels, do math that lines up by coordinate names, and run things like sum, groupby, and alignment without having to manage raw shapes by hand. It should work well with NumPy style data, support reading and writing common netCDF style scientific data, and play nicely with large datasets and parallel processing when possible.

Please make it feel polished and well documented, with tests and examples that show real world climate or science style use cases. If you need current implementation details from existing docs or best practices, look them up online first.