Fetch xz-compressed PSRA CSV files instead of LFS-stored ones

[anthonyfok]

Change fetch_psra_csv_from_model to a stub function that calls fetch_psra_csv_from_model_xz and uses fetch_psra_csv_from_model_lfs as fallback.

And maybe group the curl calls further, and try to streamline them (e.g. reduce the number of GitHub API calls) and fix #83 and #90 together.

Part of #91 and #76; related to #90, #83. Procrastinated and was stuck on it for too long, so opening this new issue, "divide and conquer", to help myself to have a clearer idea on how to tackle this.

See Curl calls analysis in our wiki for some beginning analysis of this matter.