benchmarks(agentic): drop conc=96,128 from b200 dsv4 vllm agentic sweep

Removes the two highest-concurrency points from the
tp=8/ep=8/dp-attn=true row in dsv4-fp4-b200-vllm-agentic. Sweep now
caps at conc=64 for the EP row; tp=8 plain row already caps at 16.
b300 sibling unchanged.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Signed-off-by: Cam Quilici <cjquilici@gmail.com>

Author: cquil11

.github/configs/nvidia-master.yaml

@@ -1778,7 +1778,7 @@ dsv4-fp4-b200-vllm-agentic:
       # removed for this iteration; restore from prior commits if revisiting
       # offload regressions.
       - { tp: 8, offloading: none, conc-list: [1, 2, 4, 8, 12, 16] }
-      - { tp: 8, ep: 8, dp-attn: true, offloading: none, conc-list: [12, 16, 24, 32, 48, 64, 96, 128] }
+      - { tp: 8, ep: 8, dp-attn: true, offloading: none, conc-list: [12, 16, 24, 32, 48, 64] }
 
 dsv4-fp4-b200-trt:
   image: ghcr.io#semianalysisai/trtllm-deepseek-v4:feat-deepseek_v4-9aa3715