fix(agentic-mtp): reduce chunked prefill back to 8K per scheduler

Revert per-scheduler chunked prefill from 16K to 8K to match the old
run configuration (8192*TP = 65536 total vs 16384*TP = 131072).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: seungrokj

benchmarks/single_node/agentic/dsv4_fp4_mi355x_sglang_mtp.sh

@@ -97,7 +97,8 @@ if [ "$DP_ATTENTION" = "true" ]; then
     # SGLang divides the configured chunk across DP schedulers. Use a 16K
     # per-scheduler chunk so long agentic prefill tails drain within the
     # standard 600-second warmup grace period.
-    CHUNKED_PREFILL_SIZE=$((16384 * TP))
+    # 0719
+    CHUNKED_PREFILL_SIZE=$((8192 * TP))
     PARALLEL_ARGS+=(
         --dp "$TP"
         --enable-dp-attention