Migrate the last core.constants stragglers (examples, paper validator)

Five references outside the gate scopes: the two runnable examples,
the paper's anchor validator, book/tests/test_registry.py, and a
LEGO_CELLS.md prose mention. All verified executable post-migration
(examples 03/04 run; validate_anchors: all 7 anchors match).

book/docs/LEGO_CELLS.md

@@ -50,7 +50,7 @@ The A100 ridge is `{python} A100RidgeExample.ridge_str` FLOP/byte.
 - Keep **≤ 5–8 exports** per cell when possible.
 - Use **registry paths** for shared specs — `Hardware.Cloud.H100`, `Models.Vision.ResNet50`,
   `Systems.Reliability.Gpu.mttf_hours`, `Literature.Training.MfuHigh`, and
-  `mlsysim.physics.calc_*` for architecture formulas. Reserve `mlsysim.core.constants`
+  `mlsysim.physics.calc_*` for architecture formulas. Reserve `mlsysim.core.units`
   for physics/units only (`HOURS_PER_DAY`, `BYTES_FP16`, latency stack).
 - Put `#| echo: false` as the **first line** after ` ```{python}` (required by
   `book-check-code`).

book/tests/test_registry.py

@@ -3,7 +3,7 @@
 
 import unittest
 from mlsysim import Hardware, Models
-from mlsysim.core.constants import ureg
+from mlsysim.core.units import ureg
 
 class TestMLSysRegistry(unittest.TestCase):
     def test_hardware_ridge_points(self):
@@ -21,7 +21,7 @@ class TestMLSysRegistry(unittest.TestCase):
     def test_model_size(self):
         """Test model weight storage calculations."""
         gpt3 = Models.Language.GPT3
-        from mlsysim.core.constants import BYTES_FP16, BYTES_INT4
+        from mlsysim.core.units import BYTES_FP16, BYTES_INT4
         
         # GPT-3 175B @ FP16 (2 bytes) = 350 GB
         size_fp16 = gpt3.size_in_bytes(BYTES_FP16)

mlsysim/examples/03_heterogeneous_cluster.py

@@ -7,7 +7,7 @@ from mlsysim.systems.types import Fleet, Node, NetworkFabric
 from mlsysim.hardware.registry import Hardware
 from mlsysim.models.registry import Models
 from mlsysim.solvers import DistributedModel, EconomicsModel
-from mlsysim.core.constants import Q_
+from mlsysim.core.units import Q_
 
 # 1. Define the Workload
 model = Models.Language.Llama3_8B

mlsysim/examples/04_data_wall.py

@@ -6,7 +6,7 @@ It shows how faster GPUs can actually result in lower utilization if the
 storage bandwidth cannot keep up with the compute demand.
 """
 import mlsysim
-from mlsysim.core.constants import Q_
+from mlsysim.core.units import Q_
 
 def main():
     print("Evaluating ResNet-50 Training Data Pipeline...\n")

mlsysim/paper/scripts/validate_anchors.py

@@ -17,7 +17,7 @@ project_root = Path(__file__).resolve().parent.parent.parent
 sys.path.insert(0, str(project_root))
 
 import mlsysim  # noqa: E402
-from mlsysim.core.constants import Q_  # noqa: E402
+from mlsysim.core.units import Q_  # noqa: E402
 from mlsysim.solvers import (  # noqa: E402
     DistributedModel,
     ScalingModel,