[Fix] Resolve VLM test image placeholders from the model's own chat template

[hnyls2002]

VLM tests built their image placeholder from sglang.lang.chat_template -- a frontend-only registry with no contract with the server's multimodal processors -- so models with no entry there fell back to <image> and the image was silently dropped. Tests now get the placeholder from the server or the model's own HF chat template.

---

CI States

Latest PR Test (Base): ❌ Run #30891678976
Latest PR Test (Extra): ❌ Run #30891679079

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[hnyls2002]

/rerun-test test_dp_attention.py test_skip_tokenizer_init.py test_piecewise_cuda_graph_support_1_gpu.py

[github-actions]

Results for /rerun-test test_dp_attention.py test_skip_tokenizer_init.py test_piecewise_cuda_graph_support_1_gpu.py:

🚀 2-gpu-h100 (1 test): ✅ View workflow run

cd test/ && python3 registered/dp_attn/test_dp_attention.py

🚀 1-gpu-5090 (1 test): ✅ View workflow run

cd test/ && python3 registered/tokenizer/test_skip_tokenizer_init.py

🚀 1-gpu-h100 (1 test): ✅ View workflow run

cd test/ && python3 registered/cuda_graph/piecewise/test_piecewise_cuda_graph_support_1_gpu.py