Variable.update() / Constraint.update() should validate inputs the same way add_variables() / add_constraints() do

[FBumann]

This is a tracking issue

Problem

Variable.update() and Constraint.update() currently validate inputs differently from their corresponding add_* constructors. Coord-aware checks (extra dims, mismatched coord values, reindex on equal-but-reordered) live in add_variables / add_constraints but not in the update path.

Concretely on feat/typed-update-api:

# Variable._validate_update — current
new_val = DataArray(val).broadcast_like(ref)
if not set(new_val.dims).issubset(own_dims):
    raise ValueError("Cannot assign new dimensions to existing variable.")

vs.

# add_variables — current
if isinstance(lower, DataArray):
    lower = _validate_dataarray_bounds(lower, coords)
as_dataarray(lower, coords, **kwargs)

The add_* path catches a strict superset of what update() catches: pandas Series with a wrong-named index, DataArrays with extra dims, DataArrays with shared dims but mismatched coord values, etc. update() silently broadcasts in those cases.

What should happen

After #726 lands (which renames _validate_dataarray_bounds → assert_compatible_with_coords and tightens as_dataarray's semantics), rebase Variable._validate_update and the analogous places in Constraint.update onto the same as_dataarray + assert_compatible_with_coords pipeline used by add_variables / add_constraints.

Net effect: the validation contract for changing a bound, coefficient, sign, or rhs becomes identical to the contract for creating one. Same errors, same coord coercions, no surprise behavior at the update site.

Notes

• Tracked as a follow-up to Variable.update() / Constraint.update() as canonical mutation API #727 ("Variable.update() / Constraint.update() as canonical mutation API"). The pipeline rewrite was reverted from Variable.update() / Constraint.update() as canonical mutation API #727 because it would conflict with refactor: unify as_dataarray; split broadcasting from coords validation #726 / fix: broadcast pandas/DataArray bounds in coords and preserve dim order #722 on rebase.
• Stack: fix: broadcast pandas/DataArray bounds in coords and preserve dim order #722 → fix: apply coords-as-truth rule to mask in add_variables/add_constraints #725 (merged) → refactor: unify as_dataarray; split broadcasting from coords validation #726 → this issue.
• Should land after refactor: unify as_dataarray; split broadcasting from coords validation #726 to avoid double-touching the helpers.