Update on "[ET-VK] Fused RMSNorm operator to fix fp16 overflow"

Fused RMSNorm operator that performs squaring, mean, rsqrt, and
weight scaling in a single shader dispatch. All accumulation is done
in fp32 regardless of input dtype, preventing fp16 overflow when
residual stream values exceed sqrt(65504) ≈ 256.

The Python reference impl (`rms_norm_impl`) must preserve the input
dtype — PyTorch type promotion would otherwise produce fp32 output
from fp16 inputs, and the FusePatternsPass re-trace would propagate
that incorrect dtype through the graph.

Authored by Claude.

Differential Revision: [D99841211](https://our.internmc.facebook.com/intern/diff/D99841211/)

[ghstack-poisoned]

Author: ssjia

backends/vulkan/patterns/rms_norm.py

@@ -26,11 +26,12 @@
 
 def _skip_casts(node: torch.fx.Node) -> torch.fx.Node:
     """Unwrap chains of dtype-cast nodes to find the underlying value."""
-    while isinstance(node, torch.fx.Node) and node.target in _CAST_OPS:
-        if node.args and isinstance(node.args[0], torch.fx.Node):
-            node = node.args[0]
-        else:
+    while node.target in _CAST_OPS:
+        arg0 = node.args[0] if node.args else None
+        if not isinstance(arg0, torch.fx.Node):
             break
+        node = arg0
+    # pyre-ignore[7]: node is always a Node; Pyre cannot narrow through loops
     return node