Add NaN/inf checking to stage C test

tests/unit/test_fmha_v3_stage_c_full.py

@@ -449,6 +449,8 @@ def test():
         torch.cuda.synchronize()
 
         out = c[:, :, 0].float()
+        has_nan = out.isnan().any().item()
+        has_inf = out.isinf().any().item()
         cos = torch.nn.functional.cosine_similarity(
             out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)
         ).item()
@@ -456,6 +458,7 @@ def test():
         n_tiles = n // 128
         print(f'FMHA Stage-C n={n} ({n_tiles} kv tiles): '
               f'cos {cos:.6f}  max_abs {max_abs:.4f}  '
+              f'nan={has_nan} inf={has_inf}  '
               f'{"PASS" if cos >= 0.99 else "FAIL"}')
         if cos < 0.99:
             print(f'  out[0,:4]={out[0,:4].tolist()}')