deepseek tui support issue

please check ds4 server hang when get request from deepseek tui
```
(ds4) root@salab-hpedl380g11-02:/mnt/nvme3n1/wayne/ds4/ds4# ./ds4-server --ctx 1000000 --kv-disk-dir /mnt/nvme3n1/ds4-kv --kv-disk-space-mb 2621440
ds4: Linux cuda backend set oom_score_adj=1000
ds4: CUDA backend initialized on NVIDIA L40S (sm_89)
ds4: CUDA registered 80.76 GiB model mapping for device access
ds4: CUDA preparing model tensor mappings: 80.24 GiB
ds4: CUDA q8 fp16 cache limit reached; using q8 kernels (request=64.00 MiB cached=7.94 GiB limit=8.00 GiB)
ds4: CUDA startup model preparation covered 80.76 GiB of tensor spans in 0.033s
ds4: cuda backend initialized for graph diagnostics
0712 13:00:04 ds4-server: context buffers 21302.79 MiB (ctx=1000000, backend=cuda, prefill_chunk=4096, raw_kv_rows=4352, compressed_kv_rows=250002)
ds4: CUDA using managed KV cache for ctx=1000000 (kv cache 13.17 GiB, context buffers 20.80 GiB); this may degrade performance but is needed for very large contexts
0712 13:00:04 ds4-server: KV disk cache /mnt/nvme3n1/ds4-kv (budget=2621440 MiB, cross-quant=accept, min=512, cold_max=30000, continued=10000, trim=32, align=2048, hit_half_life=21600s)
0712 13:00:04 ds4-server: listening on http://127.0.0.1:8000
0712 13:01:30 ds4-server: chat ctx=0..17642:17642 TOOLS prompt start
0712 13:01:30 ds4-server: chat ctx=0..17642:17642 TOOLS prefill chunk 0/17642 (0.0%) chunk=0.00 t/s avg=0.00 t/s 0.001s

```