Assets grid: subtract MAX_CONSUMPTION margin from the floor

With consumption now declared as `IrregSpacedGrid(n_points=N)` and
points filled at runtime from `geomspace(consumption_floor,
max_consumption, N)`, the grid clusters densely just above
`consumption_floor`. At the lowest-asset / highest-OOP-shock corner,
those near-floor consumption choices push `next_assets =
cash_on_hand - OOP - consumption` slightly below the assets grid's
old lower bound (`0` for the bare model, `-max_annual_labor_income`
when wage_params are available). Out-of-bounds interpolation of
next-period V then injects NaN, which propagates back through E[V]
and eventually fails `validate_V`. Symptom on the production solve:
`Value function at age 93 in regime 'retiree_oamc_forced_forcedout':
7317 of 207360 values are NaN`, with the `[NOTE]` showing E[V] NaN
concentrated at the lowest assets indices and the highest
hcc_transitory shock.

Subtract `MAX_CONSUMPTION` from the assets floor to give a worst-case
single-period drain margin. With 24 linspace points spanning the
wider range, the per-point density change is negligible; the dead
state and the bare-model fallback get the margin too. The asymmetry
fix is the cheapest one — no change to the consumption grid type, no
change to per-iteration parameters, no new constraints.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: hmgaudecker

src/aca_model/baseline/regimes/_common.py

@@ -262,10 +262,25 @@ def build_grids(
         sigma=1.0,
     )
 
-    assets_start = 0.0
+    # Assets-grid lower bound includes a one-period margin below the
+    # binding borrowing limit so that `next_assets = cash_on_hand - OOP -
+    # consumption` stays inside the grid even at the worst-shock × low-
+    # consumption corner. With the runtime log-spaced consumption grid
+    # `geomspace(consumption_floor, max_consumption, n_points)`, choices
+    # cluster densely just above `consumption_floor`, and at the lowest-
+    # asset/highest-OOP-shock corner those choices push `next_assets`
+    # slightly off the grid bottom — out-of-bounds interpolation of
+    # next-period V then injects NaN that propagates through E[V].
+    # Subtracting `MAX_CONSUMPTION` gives a worst-case single-period
+    # drain margin; cheap at production grid sizes (24 linspace points
+    # over the wider range).
+    assets_start = -MAX_CONSUMPTION
     if wage_params is not None and _has_required_wage_keys(wage_params=wage_params):
-        assets_start = -_compute_max_annual_labor_income(
-            wage_params=wage_params, wage_res_grid=wage_res
+        assets_start = (
+            -_compute_max_annual_labor_income(
+                wage_params=wage_params, wage_res_grid=wage_res
+            )
+            - MAX_CONSUMPTION
         )
 
     return Grids(