fix: normalize clustered solution dim order to (cluster, time) (#704)

linopy <0.8 stores the extra-timestep charge_state variable as
(time, cluster), because its 'time' coordinate (length n+1) conflicts
with the model's 'time' (length n), and 0.7 reorders the conflicting
dim to the front. Every other variable is (cluster, time). linopy >=0.8
makes coords the source of truth and is already consistent.

flixopt cannot control this from the bounds/coords it passes (0.7
reorders internally), so normalize in the solution property: transpose
'cluster' before 'time'. This is a no-op on linopy >=0.8 and for
non-clustered systems, and only touches the cluster/time axis ordering
(other dims and scalars are preserved via the ellipsis).

Also update the clustering test to access charge_state by label and
assert the now-deterministic (cluster, time) order. Unblocks the
linopy 0.8 bump in #701.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: FBumann

flixopt/structure.py

@@ -335,6 +335,8 @@ def solution(self):
         if 'time' in solution.coords:
             if not solution.indexes['time'].equals(self.flow_system.timesteps_extra):
                 solution = solution.reindex(time=self.flow_system.timesteps_extra)
+        if 'cluster' in solution.dims and 'time' in solution.dims:
+            solution = solution.transpose('cluster', 'time', ...)
         return solution
 
     @property

tests/test_math/test_clustering.py

@@ -336,12 +336,10 @@ def test_storage_cyclic_charge_discharge_pattern(self, optimize):
         discharge_fr = fs.solution['Battery(discharge)|flow_rate'].values[:, :4]
         assert_allclose(discharge_fr, [[0, 50, 0, 50], [0, 50, 0, 50]], atol=1e-5)
 
-        # Charge state: dims=(time, cluster), 5 entries (incl. final)
-        # Cyclic: SOC wraps, starting with pre-charge from previous cycle
         charge_state = fs.solution['Battery|charge_state']
-        assert charge_state.dims == ('time', 'cluster')
-        cs_c0 = charge_state.values[:5, 0]
-        cs_c1 = charge_state.values[:5, 1]
+        assert charge_state.dims == ('cluster', 'time')
+        cs_c0 = charge_state.isel(cluster=0).values[:5]
+        cs_c1 = charge_state.isel(cluster=1).values[:5]
         assert_allclose(cs_c0, [50, 50, 0, 50, 0], atol=1e-5)
         assert_allclose(cs_c1, [100, 100, 50, 100, 50], atol=1e-5)