fix: write timezone-aware coordinates in Model.to_netcdf

[MaykThewessen]

Note

AI-assisted. Claude wrote the code and tests, and ran the checks listed below. I reviewed the diff.

Changes proposed in this Pull Request

Follow-up to #897. Since that fix a model with timezone-aware coords builds and solves, but Model.to_netcdf() then raises TypeError: Cannot interpret 'datetime64[us, Europe/Amsterdam]' as a data type, because netCDF has no timezone-aware datetime type. This also breaks solve(remote=...), which goes through to_netcdf. PyPSA runs into it as soon as it allows tz-aware snapshots (PyPSA/PyPSA#983).

to_netcdf now writes tz-aware datetimes as naive UTC and stores the zone names in a _linopy_tz attribute, and read_netcdf restores them. This covers dimension indexes, MultiIndex levels, auxiliary coordinates (the flat-dimension pattern of v1), CSR-backed constraints including their aux coords, and tz-aware data variables in Model.parameters. Files without the attribute read as before.

Checklist

• AI-generated content is marked (see AGENTS.md).
• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in doc.
• Unit tests for new features were added (if applicable).
• A note for the release notes doc/release_notes.rst of the upcoming release is included.
• I consent to the release of this PR's code under the MIT license.

What was checked (AI-generated)

• New test_model_to_netcdf_tz_aware_coords in test/test_io.py, for UTC and Europe/Amsterdam over the 2020-03-29 DST change: flat index, frozen (CSR) constraint and MultiIndex level, plus a tz-aware data variable in Model.parameters. It asserts assert_model_equal and that the timezone comes back. All 10 cases fail on master (the 2 MultiIndex cases under v1 are skipped, like the existing MultiIndex fixture).
• New test_model_to_netcdf_tz_aware_aux_coords (dense and frozen, 8 cases): all fail on master with the TypeError above. Dense models already drop aux coords on read, also with naive datetimes; that is unchanged here.
• Full test suite, branch vs master: 5729 vs 5709 passed: the 18 new cases, plus 2 test_highs_keyboard_interrupt tests that failed on the master run only (flaky). The branch has no failures that master lacks; the rest come from the environment (missing oetc extras, HiGHS without SOS).
• End to end with a PyPSA network on tz-aware snapshots: n.model.to_netcdf() then read_netcdf() gives back equal snapshot indexes with the zone and an equal solution.
• Not in this PR: freezing a constraint with a MultiIndex dimension already fails on master (Grid.from_dataset, TypeError: Must pass list-like as names), so the test leaves that combination out.

🤖 Generated with Claude Code

[codspeed]

Merging this PR will not alter performance

✅ 181 untouched benchmarks
⏩ 181 skipped benchmarks¹

---

_{Comparing MaykThewessen:fix/netcdf-tz-aware-coords (a2d2846) with master (1d3c516)}

Footnotes

1. 181 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩