fix(rules): make the run reachability check avoid difficult terrain

PF1e (CRB Movement): "You can't run or charge across difficult
terrain". _execute_run checked the explicit destination against the full
cost field (which routes through difficult terrain at 2x) but descended
with avoid_difficult=True, so a destination beyond rubble was accepted
(ok=True) yet silently truncated at the obstacle. _destination_reachable
now takes the same avoid_difficult constraint (difficult cells removed
from the field; a destination whose footprint touches difficult terrain is
not a legal run endpoint either), so verdict and path agree. The test that
pinned the old silent-stop-short convention now expects
unreachable_destination; README updated.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

tests/test_aid_and_run.py

@@ -368,12 +368,13 @@ def test_run_stops_before_difficult_terrain_toward_target() -> None:
     assert eng._states[0].pos == (2, 0)
 
 
-def test_run_stops_before_difficult_terrain_to_destination() -> None:
+def test_run_beyond_difficult_terrain_to_destination_is_rejected() -> None:
     """Given: the same rubble row
-    When: hero runs to an explicit destination beyond it
-    Then: the descent stops in front of the rubble (the destination is
-    reachable in budget, but the run cannot cross — it moves up to the
-    obstacle and stops)."""
+    When: hero runs to an explicit destination ONLY reachable across it
+    Then: the destination is rejected (unreachable_destination) — the
+    reachability check uses the same avoid-difficult constraint as the
+    descent (CRB: "You can't run or charge across difficult terrain"), so
+    the verdict and the walked path agree instead of silently stopping short."""
 
     def policy(_engine: CombatEngine, state: CombatantState) -> tuple[Action, ...]:
         if state.combatant.id == "hero":
@@ -382,9 +383,28 @@ def test_run_stops_before_difficult_terrain_to_destination() -> None:
 
     grid = _rubble_row_grid(size=8, row=3)
     hero = make_state(_combatant("hero", initiative_mod=2), "players", (0, 0))
-    # The default foe sits at (0, 1), adjacent to the start: the run walks
-    # out of its threatened disk and provokes one (missed) AoO.
-    eng = _duo(policy, [10, 5, 1, 1], round_cap=1, hero=hero, grid=grid)
+    eng = _duo(policy, [10, 5], round_cap=1, hero=hero, grid=grid)
+    result = eng.run()
+    assert eng._states[0].pos == (0, 0)
+    assert "round 1 hero: run rejected (unreachable_destination)" in result.transcript
+
+
+def test_run_destination_in_front_of_difficult_terrain_is_reachable() -> None:
+    """Given: the same rubble row
+    When: hero runs to an explicit destination that avoids the rubble
+    Then: the run honours it (the field routes around difficult cells, the
+    budget is generous — the verdict is positive and the path arrives)."""
+
+    def policy(_engine: CombatEngine, state: CombatantState) -> tuple[Action, ...]:
+        if state.combatant.id == "hero":
+            return (Action(kind="run", destination=(2, 0)),)
+        return _WAIT
+
+    grid = _rubble_row_grid(size=8, row=3)
+    hero = make_state(_combatant("hero", initiative_mod=2), "players", (0, 0))
+    # The default foe at (0, 1) threatens the start: the run provokes one
+    # (missed) AoO on the first step — scripted as a natural 1.
+    eng = _duo(policy, [10, 5, 1], round_cap=1, hero=hero, grid=grid)
     result = eng.run()
     assert eng._states[0].pos == (2, 0)
     assert "round 1 hero: run" in result.transcript