Remove pygrib as a dependency; manually parse coordinate reference system from eccodes keys

[blaylockbk]

I was having issues with pygrib after its latest release (something to do with numpy 2+ releases). Pygrib was working fine on my personal computer, but I was getting segmentation faults in my GitHub Actions tests. I decided to drop pygrib as a dependency since I was only using it to parse the coordinate reference system (CRS) when loading the data into xarray. It's not too painful to just parse the CRS info (in CF-compliant form) from what cfgrib can provide. I wish cfgrib would handle the CF grid specification itself (ecmwf/cfgrib#251) but I can let Herbie do it for now. Maybe I'll contribute to cfgrib if I have some free time.

• Remove pygrib as a dependency.
• Manually parse coordinate reference system (CRS) details for pyproj, then convert to CF convention.
• Am I getting enough data needed to parse crs from all the models Herbie supports???
• Updated pinned dependencies (e.g., numpy>=2.2.1). I don't know if this is a good thing yet. I could relax if people complain. It may make using xesmf un-usable until development on numba catches up to the latest numpy releases.

---

Also, fixed the GitHub Actions tests!

[conbrad]

Thank you for addressing this, we ran into this issue too running on Linux containers and would have contributed but were just as puzzled. Likely a memory issue in pygrib so I like this solution!