Tests: actually gate FP8 use based on capability

tests/test_triton_scaled_mm.py

@@ -37,9 +37,9 @@ def get_8bit_types():
     capability = major * 10 + minor
     supports_fp8 = capability >= 89
 
-    if torch.version.hip is not None:
+    if supports_fp8 and torch.version.hip is not None:
         types.append(torch.float8_e4m3fnuz)
-    elif torch.version.cuda is not None and torch.cuda.is_available():
+    elif supports_fp8 and torch.version.cuda is not None and torch.cuda.is_available():
         types.append(torch.float8_e4m3fn)
     return types