[ET-VK][ez] Fix partitioner logic of finding keepdim arg of reduce ops (#13598)

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at
bottom):
* #13597
* #13596
* #13595
* #13594
* #13593
* #13600
* #13599
* __->__ #13598

Title says it all. For reduce ops, their signature are not all alike so
some extra legwork needs to be done to identify specific arguments that
need to be checked.

Also included a small update to partitioner logging to improve
debuggability.

Differential Revision:
[D80741737](https://our.internmc.facebook.com/intern/diff/D80741737/)

---------

Co-authored-by: ssjia <[email protected]>

Author: SS-JIA

backends/vulkan/op_registry.py

@@ -397,14 +397,17 @@ def check_reduce_node(node: torch.fx.Node) -> bool:
                 # If we can't get memory layout information, we'll assume the dims aren't packed
                 pass
 
-        keepdim = node.args[2]
-        if isinstance(keepdim, bool) and not keepdim:
+        def try_find_keepdim_arg(node: torch.fx.Node) -> bool:
+            for arg in node.args:
+                if isinstance(arg, bool):
+                    return arg
+
+            # Assume false by default
             return False
 
-        if len(node.args) > 2:
-            keepdim = node.args[2]
-            if isinstance(keepdim, bool) and not keepdim:
-                return False
+        keepdim = try_find_keepdim_arg(node)
+        if isinstance(keepdim, bool) and not keepdim:
+            return False
 
         return True
 

backends/vulkan/partitioner/vulkan_partitioner.py

@@ -204,7 +204,7 @@ def is_in_local_scalar_dense_chain(self, node: torch.fx.Node) -> Tuple[bool, boo
     def log_skip(self, node: torch.fx.Node, reason: str) -> None:
         if node.op == "call_function":
             logger.info(
-                f"[Vulkan Partitioner] Due to [{reason}], skipping {node.format_node()}"
+                f"[Vulkan Partitioner] Due to [{reason}], skipping {utils.node_io_str(node)}"
             )
 
     def is_node_supported(

backends/vulkan/utils.py

@@ -1059,6 +1059,8 @@ def get_node_val_str(node: torch.fx.Node) -> str:
         assert isinstance(node.meta["val"], (list, tuple))
         return f"[{', '.join(get_tensor_val_str(t) for t in node.meta['val'])}]"
     else:
+        if "val" not in node.meta:
+            return str(node)
         return str(node.meta["val"])