fix: add_variables ignoring coords for DataArray bounds

[FBumann]

Summary

Two bugs in add_variables when bounds are DataArrays and coords is provided. Both lower and upper are affected identically.

This is a focused extract from #551 — bug fixes only, no as_dataarray refactoring.

Bug 1: DataArray coords silently ignored → NaN bounds

lower = xr.DataArray([0, 0, 0], dims=["time"], coords={"time": [0, 1, 2]})
var = m.add_variables(lower=lower, coords=[pd.RangeIndex(5, name="time")], name="x")

# master:  var.data.lower.values → [0, 0, 0, NaN, NaN]  (silent NaN)
# fixed:   ValueError: Coordinates for dimension 'time' do not match

lower = xr.DataArray([0, 0, 0], dims=["time"], coords={"time": [10, 11, 12]})
var = m.add_variables(lower=lower, coords=[pd.RangeIndex(5, name="time")], name="y")

# master:  var.data.lower.values → [NaN, NaN, NaN, NaN, NaN, 0, 0, 0]  shape (8,)!
# fixed:   ValueError: Coordinates for dimension 'time' do not match

The fix validates DataArray bounds against coords: mismatched or extra dims raise ValueError, missing dims are broadcast via expand_dims, reordered coords (same values) are reindexed.

Bug 2: Dict coords crash for all input types

m.add_variables(lower=0, coords={"x": [0, 1, 2]}, name="x")

# master:  CoordinateValidationError (dims not inferred from dict keys for scalar upper=inf)
# fixed:   works — shape (3,)

Fixed by inferring dims from dict keys in as_dataarray and numpy_to_dataarray for scalar/0-dim inputs.

Changes

File	Change
linopy/model.py	_validate_dataarray_bounds + _coords_to_dict helpers, called in add_variables
linopy/common.py	Infer dims from dict keys for scalars/0-dim arrays
test/test_common.py	45 new parameterized tests (16 fail on master, all pass here)

No existing tests were modified or removed — all changes to test/test_common.py are purely additive. All 32 existing test functions from master are preserved.

Test results: master vs this branch

Category	#	Master	Fixed
All types × dict-coords	9	Crash (CoordinateValidationError)	Pass
DataArray coord mismatch/extra dims	5	Silent NaN (wrong bounds, no error)	ValueError
Reordered coords (same values)	1	Pass (xarray alignment, no NaN)	Explicit reindex
Reordered coords (different values)	1	Silent NaN (wrong bounds, no error)	ValueError
All types × seq-coords, mixed combos, broadcast, inferred coords, string/datetime, MultiIndex	29	Pass	Pass
Total	45	29 pass, 16 fail	45 pass

Full test suite: 2570+ passed, 236 skipped.

🤖 Generated with Claude Code

[FBumann]

@FabianHofmann Sorry for all the branches. I really wanted to make the fix more universal and establish a distinction between converting to a dataarray and converting to a dataarray and validate, but i hot lost in the process.
This PR now soley focuses on add_variables() bugs and inconsistencies.

[FabianHofmann]

very good! thanks as always @FBumann ; the tests look very sensible. I think they functionally overlap a little with the tests in test_variable.py where they should be moved as well. So I would be fine in keeping them but move them there. otherwise I see this ready to merge

[FabianHofmann]

I guess this is something we need to keep in mind for #591 where this should be disallowed

[FBumann]

Yes!

[FBumann]

@FabianHofmann all set to merge