fix(cost_calculator): narrow exception catch to model-not-found only

Replace bare 'except Exception' with a targeted check for the
'isn't mapped yet' message, re-raising all other exceptions so
real pricing failures remain visible. Adds noqa: BLE001 to satisfy
the ruff strict budget.

Addresses review feedback from greptile and veria-ai bots.

litellm/cost_calculator.py

@@ -606,7 +606,9 @@ def cost_per_token(
                 service_tier=service_tier,
                 data_residency=data_residency,
             )
-        except Exception:
+        except Exception as e:  # noqa: BLE001
+            if "isn't mapped yet" not in str(e):
+                raise
             verbose_logger.debug(
                 "cost_per_token: unknown model=%s, custom_llm_provider=openai. Returning 0.0, 0.0",
                 model,