Clarify in docs that the deferred layer records setter calls

Per review discussion: keep the "deferred" name but make the docstrings state
plainly that the mechanism is call recording -- the class records the problem
and defers building the device model, it does not build anything itself.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Signed-off-by: Ramakrishna Prabhu <ramakrishnap@nvidia.com>

Author: ramakrishnap-nv

python/cuopt/cuopt/routing/_deferred.py

@@ -1,11 +1,14 @@
 # SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-"""Store-then-build (deferred build) layer for the routing DataModel.
-
-The public DataModel records its setter calls on the host and builds the device
-(Cython) model only transiently -- at solve, or when a getter is queried -- by
-replaying the recorded calls. Design points:
+"""Deferred-build layer for the routing DataModel: record setter calls, build on demand.
+
+The public DataModel **records** each setter call on the host instead of applying
+it to the GPU immediately. The class does not build anything itself -- it stores
+the problem and defers construction of the device (Cython) model until it is
+actually needed (at solve, or when a getter is queried), at which point it builds
+transiently by replaying the recorded calls onto the wrapper. "Deferred" refers
+to that deferred build; the mechanism is call recording. Design points:
 
   * **Host-resident IR.** Host inputs (numpy/pandas) are copied to a numpy array
     the DataModel owns at record time, so the recorded calls form a serializable,

python/cuopt/cuopt/routing/vehicle_routing.py

@@ -51,9 +51,9 @@ class DataModel(_DeferredDataModel):
         host (numpy/pandas) inputs are copied to the device for the local
         solve. Python lists and tuples are not supported.
 
-      - Inputs are recorded and the device model is built lazily, so
-        solver-side (C++) validation and dtype-cast warnings surface when the
-        model is built -- at ``Solve`` -- rather than at the individual setter
+      - Inputs are recorded and the device model is built on demand (deferred
+        to ``Solve``), so solver-side (C++) validation and dtype-cast warnings
+        surface when the model is built rather than at the individual setter
         call. Structural checks (matrix shape, array sizes, value ranges) are
         still validated eagerly at the setter.