fix: truncate excess scenarios instead of raising error

When the LLM generates more scenarios than requested, truncate to the
requested count instead of raising a ValueError. This makes scenario
generation more robust since LLMs occasionally overshoot the count.
Still raises an error when too few scenarios are generated.

https://claude.ai/code/session_017Y9KNNQX2RyVWnqpj3A4hh

Author: claude

src/art/mcp/generate_scenarios.py

@@ -201,9 +201,14 @@ async def generate_scenarios(
         scenarios = result if isinstance(result, list) else list(result.values())[0]
 
     # Validate count
-    if len(scenarios) != num_scenarios:
+    if len(scenarios) < num_scenarios:
         err(f"Expected {num_scenarios} scenarios, got {len(scenarios)}.")
         raise ValueError(f"Expected {num_scenarios} scenarios, got {len(scenarios)}")
+    elif len(scenarios) > num_scenarios:
+        ok(
+            f"Expected {num_scenarios} scenarios, got {len(scenarios)}. Truncating to {num_scenarios}."
+        )
+        scenarios = scenarios[:num_scenarios]
 
     ok(f"Parsed {len(scenarios)} scenario(s) successfully.")