Three levels: step unpacks only the first two levels; README multipatch nits

step() did 'c, f = self._levels', which raises for three levels before any
launch (root review). The child level is driven from the fine hops. Other
_levels uses are per-level loops or explicit indices. README multipatch
example now defines rest locally and states coarse/fine NaN separately.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

README.md

@@ -145,13 +145,15 @@ Scope and limits:
 
   ```python
   cfg = LBMConfig(nx=32, ny=16, viscosity=.1)
+  rest = np.asarray(compute_feq_d2q9(1.0, 0.0, 0.0), np.float32)[:, None, None]
   with CudaContext(0) as ctx, FixedPatchStepper(
           cfg, ((4, 4, 12, 12), (16, 4, 24, 12)), context=ctx,
           fine_block=(4, 4), base_block=(4, 4)) as run:
       run.initialize(np.broadcast_to(rest, (9, *run.patch.coarse_shape)),
                      np.broadcast_to(rest, (9, *run.patch.fine_shape)))  # (9, 32, 64)
       run.step(32)
-      coarse_g, fine_g = run.download()  # NaN outside every patch's active cells
+      coarse_g, fine_g = run.download()  # coarse: NaN on every covered cell;
+                                         # fine: NaN on every non-ACTIVE cell
   ```
 - Not available: per-link surface exchange/force readout in blocked mode,
   checkpoints, moving bodies, bodies/Jets with several patches, dynamic refinement, `Simulation`

src/CelerisLab/lbm/refinement_stepper.py

@@ -496,7 +496,7 @@ class FixedPatchStepper:
             raise ValueError("n must be a nonnegative integer")
         with self._lock, self.ctx.activate():
             self._require("ready")
-            c, f = self._levels
+            c, f = self._levels[:2]  # a nested child (third level) runs inside the fine hops
             p = self.patch
             kw = dict(block=self._block, stream=self._stream)
             try: