Add Variable.reindex (mirroring LinearExpression.reindex)

[FabianHofmann]

Note

This issue was drafted by AI (Claude).

Describe the feature you'd like to see

LinearExpression has a .reindex method, but Variable does not — so a Variable cannot be aligned to a new coordinate set directly. To reindex a variable today you have to round-trip through .to_linexpr() (or reindex a derived DataArray such as .isnull() separately), which is awkward when you just want to align the variable itself.

import pandas as pd
from linopy import Model

m = Model()
names = pd.Index(["a", "b"], name="name")
v = m.add_variables(coords=[names], name="x")

hasattr(v, "reindex")                 # False
hasattr(v.to_linexpr(), "reindex")    # True

v.reindex(name=pd.Index(["a", "b", "c"], name="name"))   # AttributeError

A Variable.reindex(...) mirroring LinearExpression.reindex (and the underlying xarray semantics) would let callers align a variable to a target index in a single expression, without converting to a LinearExpression first.

Masking behavior

A masked entry already carries variable label -1, the same sentinel reindex uses for a freshly-introduced coordinate — so masking falls out with a single notion of "absent":

1. New coordinates fill as masked/null, not with a value (fill_value doesn't apply to variable labels — there's no variable to fill).
2. Existing masks are preserved; subsetting just drops entries.
3. isnull() stays mask-aware: it reports True for both pre-existing masks and newly-missing coords, with no special-casing.

v.reindex(name=["a", "b", "c"]).isnull()   # [False, True, True]
#                                              pre-existing mask ^   ^ new coord

The value is a correct, mask-aware null array at the Variable level — before to_linexpr(), which is exactly where one wants to branch on availability. Constant fills (e.g. 1 for "always on") remain a LinearExpression concern (to_linexpr().reindex(...).where(...)); Variable.reindex stays pure: variables or masked, nothing else.

[FBumann]

This is already part of the v1 semantics PR. But we can extract it into its own or. Alongside a few other things maybe

[FabianHofmann]

ah great, don't mind. v1 is the next big thing to be worked on, so fine if we leave it there