[AMD][Kernel][Bugfix] Cast offsets tensor bn to tl.int64 to avoid GPU segfault#23692

gshtras merged 5 commits intovllm-project:mainvllm-project/vllm:mainfrom

rasmith:ransmith_fix_prefix_prefill_segvrasmith/vllm:ransmith_fix_prefix_prefill_segvCopy head branch name to clipboard