v0.2.2

In this release (v0.2.2):

• Major bugs involving GPU-calculations on different CUDA, Julia versions are resolved.
• GPU availability is auto-detected and cross-sections are by default performed using GPU if possible, CPU otherwise. (These can still be overridden by passing architecture=XPU() to make_hitran_model or make_interpolation_model)
• Users are warned when trying to use CEF's other than HumlicekWeidemann32SDErrorFunction, the GPU-compatible and tested CEF. This CEF is now the default when performing cross-section calculations.
• Travis CI is replaced entirely with Github Actions. All tests are performed on CPU and GPU – CPU tests on Github server and GPU tests on tofu.gps.caltech.edu.