Add n_wage_res_batch_size: splay wage_res shock axis

Wires `RouwenhorstAR1Process.batch_size` through `GridConfig` so the
wage-residual stochastic productmap can be split with an inner Python
loop. At `n_wage_res_batch_size=1` the per-target Q_and_F intermediate
shrinks by `n_wage_res_gridpoints` (5), bringing the kernel under 80 GB
for the ACA-overlay nongroup_nomc_* regimes where the unsplayed kernel
hit 144 GB.

`n_pref_type_batch_size` remains a no-op pending separate splay wiring.

🤖 Generated with Claude Code
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: hmgaudecker

src/aca_model/baseline/regimes/_common.py

@@ -242,6 +242,7 @@ def build_grids(
         rho=_WAGE_RHO,
         sigma=(1.0 - _WAGE_RHO**2) ** 0.5,
         mu=0.0,
+        batch_size=grid_config.n_wage_res_batch_size,
     )
     hcc_persistent = get_hcc_persistent_shock(grid_config=grid_config)
     hcc_transitory = NormalIIDProcess(

src/aca_model/config.py

@@ -42,6 +42,14 @@ class GridConfig:
     # all pref-types. Defaults to `0` — the production overrides set it
     # to `1` on hardware where the unsplayed kernel doesn't fit.
     n_pref_type_batch_size: int = 0
+    # `batch_size` on the `wage_res` stochastic shock process: chunked
+    # productmap stride along the wage-residual stoch axis inside Q_and_F.
+    # `1` shrinks the per-target Q intermediate by `n_wage_res_gridpoints`
+    # at the cost of an inner Python loop; `0` lets the productmap span
+    # the full axis. Defaults to `0` — production overrides set it to `1`
+    # on hardware where the ACA-overlay per-cell DAG blows the kernel's
+    # compile-time working set past device HBM.
+    n_wage_res_batch_size: int = 0
     # Per-device chunk size for the simulate-side per-subject dispatch,
     # keyed by `log_level`. Empty → 0 (no chunking) for every level.
     # `log_level="off"` skips `validate_V` and its forced host-sync, which