refactor: move as_dataarray_in_coords to common.py (#723 part 1)

[FBumann]

First slice of #723. Stacked on top of #722 — set base to fix/bounds-coords-broadcast; rebase / change base to master once #722 merges.

What changes

Pure relocation, no behavior change.

• _as_dataarray_in_coords and its two helpers (_coords_to_dict, _named_pandas_to_dataarray) move from linopy/model.py to linopy/common.py, alongside the as_dataarray they parallel.
• The main helper is renamed as_dataarray_in_coords (no leading underscore) so other modules can import it when migrating call sites.
• add_variables imports + uses the relocated function; no other call sites change in this PR.

Diff is 134 / -132 lines, almost all of which is the relocation.

Audit of remaining as_dataarray(arr, coords=...) call sites

Not migrated in this PR. Per-site notes:

model.py

• add_constraints: as_dataarray(sign) and as_dataarray(rhs) pass no coords — strict semantics don't apply. No migration.
• add_variables / add_constraints mask: as_dataarray(mask, coords=data.coords, dims=data.dims).astype(bool) followed by broadcast_mask(mask, data.labels). The strict helper would handle extra-dim and expand_dims steps, but broadcast_mask also emits a FutureWarning for sparse-coord NaNs. Migration needs that warning path threaded through (or moved into the helper). Deferred — follow-up commit.

expressions.py (lines 584, 613, 1105, 1668, 2154)

• Arithmetic ops pass coords=self.coords, dims=self.coord_dims to label scalar/array inputs, then call _align_constant for the actual alignment with join/reindex. The strict helper would raise on inputs carrying an extra dim that _align_constant would otherwise broadcast across. Higher risk — defer until we have a test that pins the desired behavior for LinearExpression + DataArray_with_extra_dim.

variables.py (line 330)

• Same shape as the expression arithmetic sites. Deferred with them.

Follow-up scope

A second PR after this lands:

1. Migrate the mask paths (move the broadcast_mask NaN-warning logic into as_dataarray_in_coords or compose them explicitly).
2. Decide whether expression arithmetic should validate against self.coords. Likely needs a tightened semantic in the v1 convention work (feat: v1 semantic convention #717) rather than an unannounced behavior change here.
3. Benchmark add_variables and add_constraints on a representative model (PyPSA earth?) and compare before/after.

Test plan

• test/test_variable.py — 54/54 pass (incl. the MultiIndex + reindex coverage from fix: broadcast pandas/DataArray bounds in coords and preserve dim order #722)
• Broader sweep: test/test_model.py, test/test_linear_expression.py, test/test_constraint.py, test/test_constraints.py, test/test_piecewise_constraints.py, test/test_variable_assignment.py, test/test_variables.py, test/test_quadratic_expression.py, test/test_scalar_*.py, test/test_sos_constraints.py — 838 pass / 4 skipped, no regressions.
• Pre-commit (ruff/format/blackdoc/codespell) clean.

🤖 Generated with Claude Code

[FBumann]

Folded into #722 — the relocation + the cleaner stack(level=…) form for _named_pandas_to_dataarray are now part of that PR (commits bca89e7 and b28f3df). Closing as superseded.