fix(compiler): default cluster_dims=(1,1,1) for backends that omit it

Backends without cluster launch support (e.g. hcu, mthreads) do not emit
cluster_dims in the compiled-kernel metadata JSON, so the KernelMetadata
namedtuple built here lacks that field. torch._inductor reads
kernel.metadata.cluster_dims unconditionally when a kernel carries metadata,
raising AttributeError during KV-cache profiling warmup (vllm 0.24.0 on
Hygon DTK 26.04). Default it to (1,1,1) — the same value vendor triton and
the iluvatar overlay already use — instead of patching torch/inductor per
backend in each vendor adapter.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: tengqm

python/triton/compiler/compiler.py

@@ -434,7 +434,11 @@ def __init__(self, src, metadata_group, hash):
         target = metadata['target']
         metadata['target'] = GPUTarget(target['backend'], target['arch'], target['warp_size'])
         # Restore tuple-typed metadata fields serialized as JSON arrays.
-        cluster_dims = metadata.get("cluster_dims")
+        # Backends that do not support cluster launches (e.g. hcu, mthreads)
+        # omit cluster_dims entirely; torch._inductor reads it unconditionally
+        # when the kernel carries metadata, so default it here rather than
+        # letting every such backend patch torch/inductor in its own adapter.
+        cluster_dims = metadata.setdefault("cluster_dims", (1, 1, 1))
         if isinstance(cluster_dims, list):
             metadata["cluster_dims"] = tuple(cluster_dims)
         KernelMetadata = namedtuple('KernelMetadata', sorted(list(metadata.keys())))