[ROCm][AITER][Bugfix] Switch AITER to use PIECEWISE_AND_FULL compilation (#25104)

Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>

vllm/v1/attention/backends/rocm_aiter_fa.py

@@ -232,7 +232,7 @@ class AiterFlashAttentionMetadata:
 
 class AiterFlashAttentionMetadataBuilder(
         AttentionMetadataBuilder[AiterFlashAttentionMetadata]):
-    cudagraph_support = AttentionCGSupport.ALWAYS
+    cudagraph_support = AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
 
     def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str],
                  vllm_config: VllmConfig, device: torch.device):