feat(llm): support Ling-3.0-tiny with vLLM and SGLang

[Minamiyama]

Enable vLLM and SGLang for the existing Ling-3.0-tiny BF16, FP8 and INT4 variants from Hugging Face and ModelScope.

• Add backend-specific virtualenv requirements: vllm==0.29.0 and sglang==0.5.19, matching Ling-3.0-flash.
• Reuse the existing BailingMoeV3ForCausalLM backend adapters.
• Add regression tests for engine discovery, backend matching, launch-class selection and dependency resolution.
• Regenerate the English model documentation.

[qinxuye]

LGTM. Checked engine discovery, launch-class selection, dependency resolution and reuse of the existing Ling-3.0 backend adapters for BF16/FP8/INT4. All 60 Ling-3.0 tiny/flash regression tests passed locally. GPU model inference was not run locally.