fix(OOM): max-num-batched-tokens 4096→256 — profiling Q-tiling also OOMs

flash_attn skip worked but Q-tiling fallback still OOMs at 4096 tokens.
K tensor: [28_heads, 4096, 256] float32 = 112MB per layer slice.
At 256 tokens: [28, 256, 256] = 7MB — safe for profiling.

This only affects profiling dummy batch size. Real inference chunked
prefill still processes up to max_model_len tokens.

computility-run.yaml

@@ -19,7 +19,7 @@ command:
     - --disable-log-requests
     - --disable-frontend-multiprocessing
     - --max-num-batched-tokens
-    - '4096'
+    - '256'
     - --enable-chunked-prefill
     - --max-seq-len-to-capture
     - '32768'