ENH: cache downloaded atmosphere netCDF datasets (#654)

[thatrandomasiandev]

Summary

• Cache Forecast/Ensemble (and Reanalysis remote URLs) location-time profiles as compact .nc files under ~/.rocketpy_cache/atmosphere (override with ROCKETPY_CACHE).
• Cache Windy JSON responses on the same path; second identical set_atmospheric_model calls reuse disk.
• Add no_cache=False to Environment.set_atmospheric_model (and Windy processing) so callers can force a fresh download without breaking existing callers.

Notes / TODOs

OPeNDAP "Best" aggregations are virtual catalogs, not downloadable files, so this PR caches the extracted profiles RocketPy already materializes after the first fetch (same end-user outcome: reruns skip the network). Remaining follow-ups:

• Open-Meteo / Meteomatics JSON response caching
• Optional full-file download cache when a remote source is a concrete HTTP .nc (non-OPeNDAP)

Fixes #654

Test plan

• PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 pytest tests/unit/environment/test_atmosphere_cache.py
• Confirm Forecast shortcut reuses disk on rerun with same lat/lon/date
• Confirm no_cache=True forces re-download

[Gui-FernandesBR]

Not ready to merge yet. Three items, the third being the one I care most about.

1. Three of this PR's own new tests fail on all six platforms.

FAILED tests/unit/environment/test_environment.py::test_set_atmospheric_model_normalizes_shortcut_case_for_forecast
FAILED tests/unit/environment/test_environment.py::test_forecast_shortcut_and_dictionary_are_case_insensitive[AIGFS]
FAILED tests/unit/environment/test_environment.py::test_forecast_shortcut_and_dictionary_are_case_insensitive[HRRR]

IndexError: too many indices for array: array is 0-dimensional, but 2 were indexed
  rocketpy/environment/environment.py:1637

2. pylint fails (exit 8):

rocketpy/environment/atmosphere_cache.py:158:0: R0915: Too many statements (30/25)
rocketpy/environment/atmosphere_cache.py:270:0: R0915: Too many statements (35/25)
rocketpy/environment/environment.py:1237:4: R0912: Too many branches (28/25)

Please split those into smaller functions rather than adding # pylint: disable.

3. Cache invalidation — the design point I want resolved before this lands.

The cache is enabled by default, and the key contains neither a TTL nor the forecast model run/cycle. So a user who runs a GFS forecast today and again tomorrow silently gets yesterday's forecast, with no indication anything is stale. For forecast products that is a wrong-answer bug, not a stale-file annoyance, and it is the kind of thing that is very hard for a user to notice.

What I would want:

• the model run / cycle timestamp as part of the cache key, or a TTL, so a new cycle can never be served from an old file;
• caching off by default until we have confidence in the invalidation;
• and it is worth distinguishing product classes: reanalysis/archive data (ERA5 and friends) is immutable and can be cached indefinitely, forecasts cannot.

Happy to revisit once those are addressed.

[codecov]

Codecov Report

❌ Patch coverage is 76.52812% with 96 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.41%. Comparing base (0923d8a) to head (81406da).

Files with missing lines	Patch %	Lines
rocketpy/environment/atmosphere_cache.py	77.15%	69 Missing ⚠️
rocketpy/environment/environment.py	74.28%	27 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1137      +/-   ##
===========================================
- Coverage    91.51%   91.41%   -0.10%     
===========================================
  Files          131      132       +1     
  Lines        17733    18133     +400     
===========================================
+ Hits         16229    16577     +348     
- Misses        1504     1556      +52     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.