Open the GFS and GEFS virtual quickstarts from the catalog

[aldenks]

Draft. Must not merge before the production flip lands and deploys — see Blocking below.

Replaces the temporary direct-Icechunk cells in both virtual quickstarts with the concise dynamical_catalog.open(...) form reading the production catalog. No staging URL override.

Each notebook's opening cell already carried a TEMPORARY comment spelling out exactly this replacement; this applies it.

Changes

noaa-gfs-forecast-analysis-virtual.ipynb

import dynamical_catalog

forecast = dynamical_catalog.open("noaa-gfs-forecast-virtual", chunks=None)
analysis = dynamical_catalog.open("noaa-gfs-analysis-virtual", chunks=None)
analysis_pressure = dynamical_catalog.open("noaa-gfs-analysis-virtual", group="pressure_level", chunks=None)
forecast

noaa-gefs-forecast-analysis-virtual.ipynb likewise, opening its four datasets and two groups.

Per the same comment's instruction and the house style in noaa-hrrr-analysis-virtual.ipynb, the remaining imports move to the first cell that uses them: in the GFS notebook cartopy.crs to the 500 hPa cell, numpy to the forecast-error cell and matplotlib.pyplot to the animation cell. The GEFS notebook already imported matplotlib.pyplot and pandas at first use, so only its open cell changes. icechunk and xarray are no longer imported by either notebook.

Net: 15 insertions, 63 deletions.

Blocking

These six datasets are staging-only in the production STAC catalog today, so dynamical_catalog.open(...) cannot resolve them yet. CI will be red on this PR until all of the following happen, in order:

1. The production flip merges (dynamical-stac PR flipping the six virtuals to production).
2. That merge uploads the regenerated stac/ tree to the production catalog.
3. The stored outputs here are regenerated against the live catalog.

Step 3 is deliberately not done yet: the outputs currently committed were produced by the direct-Icechunk cells, and re-running before the flip would only reproduce the failure. Once the catalog serves these ids, re-run both notebooks and push the refreshed outputs before marking this ready.

Ordering is also why this is a draft rather than an automerge candidate: merging it early would break two notebooks on main.