test: DSv4-Pro B300 - retry-only subset (7 jobs)

Run 25831988077 hit a transient cudaErrorDevicesUnavailable at vLLM
worker startup on 7 of 36 configs (low-conc tp=8 jobs that landed on
runners while a previous job was still releasing GPUs). Other 29
configs verified clean. This commit narrows the sweep to just those
7 for a quick retry.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: cquil11

.github/configs/nvidia-master.yaml

@@ -2844,10 +2844,11 @@ dsv4-fp4-b300-vllm:
     agentic-coding:
     - duration: 1800
       search-space:
-      - { tp: 4, offloading: none, conc-list: [1, 2, 4, 8, 16, 32, 48, 64, 70] }
-      - { tp: 8, offloading: none, conc-list: [1, 2, 4, 8, 16, 32, 48, 64, 70] }
-      - { tp: 4, ep: 4, dp-attn: true, offloading: none, conc-list: [1, 2, 4, 8, 16, 32, 48, 64, 70] }
-      - { tp: 8, ep: 8, dp-attn: true, offloading: none, conc-list: [1, 2, 4, 8, 16, 32, 48, 64, 70] }
+      # Retry-only subset: re-running the 7 jobs that hit a transient
+      # cudaErrorDevicesUnavailable at vLLM worker startup in run 25831988077.
+      # The other 29 configs verified clean from that run.
+      - { tp: 8, offloading: none, conc-list: [1, 2, 8, 16] }
+      - { tp: 8, ep: 8, dp-attn: true, offloading: none, conc-list: [1, 4, 8] }
 
 dsv4-fp4-b300-trt:
   image: ghcr.io#semianalysisai/trtllm-deepseek-v4:feat-deepseek_v4-9aa3715