Drop numa-bind on the DEP2 prefill workers

Two DEP2 prefill workers share a node (GPUs 0-1 and 2-3) under CPU affinity set by the launcher, so vLLM's automatic
GPU-to-NUMA detection refuses to run and the prefill engines exit; an explicit numa-bind-nodes list cannot fit both
workers. Matches the other DEP2 vLLM recipes, which do not set numa-bind.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Author: Juntian777

inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/dsv41flash/vllm/gb300-fp4/agentx/disagg-variants.yaml

@@ -126,7 +126,6 @@ base:
         max-num-batched-tokens: 16384
         long-prefill-token-threshold: 8192
         prefix-cache-retention-interval: 32768
-        numa-bind: true
     decode:
       nodes: 4
       workers: 1