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

hcu and mthreads backends do not support cluster launch, so their
compiled kernel metadata omits cluster_dims. The KernelMetadata
namedtuple is then built without that key, and torch 2.9.0's
make_launcher reads kernel.metadata.cluster_dims unconditionally,
raising AttributeError during KV-cache profiling warmup.

Default the key to (1,1,1) before building the namedtuple, matching
what the vendor triton 3.5.1 compiler and the iluvatar overlay already
use — instead of patching torch/inductor per backend in each vendor
adapter.

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())))