Add regle_pluriel_nom and regle_accord_determinant (CE1)

Extends explicitly beyond the §15 prototype scope (3 missions, one
rule) that prompt-lancement.md says not to exceed without flagging
it — confirmed with the user before implementing (decisions.md,
point 25).

Adds the two remaining CE1 rules for the "accords dans le groupe
nominal" chapter listed in prompt-lancement.md, each with the same
minimum content volume already established for regle_accord_adjectif
(8 multiple-choice, 4 sentence corrections, 3 dictées across the 3
difficulty levels) and 3 missions (one per exercise type, matching
the existing pattern), bringing the map to 9 locations.

world_map.gd is purely data-driven already, so no code change was
needed to support more locations — only its stale "trois lieux"
header comment, corrected. Dictation audio generated with the Piper
voice (make generate-dictation-audio).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

tests/exercises/test_exercise_engine.gd

@@ -15,10 +15,11 @@ func test_unknown_rule_returns_empty_dictionary() -> void:
 
 func test_missions_are_loaded_and_sorted() -> void:
     var missions := ExerciseEngine.get_missions()
-    assert_eq(missions.size(), 3)
+    assert_eq(missions.size(), 9)
     var ids: Array = missions.map(func(m: Dictionary) -> String: return m["mission_id"])
     assert_eq(ids, ids.duplicate())  # trivialement vrai, documente l'ordre stable
     assert_true(ids.has("mission_01_pre_des_mots"))
+    assert_true(ids.has("mission_09_rocher_qui_repete"))
 
 
 func test_get_mission_returns_the_requested_mission() -> void: