Correct the sampling params

[ykhrustalev]

Problem

Inference examples across the docs used incorrect or missing sampling parameters (e.g., temperature=0.7), leading users to suboptimal generation quality.

Upstream HF Model Card Recommended Settings

| Model Family               | temp | top_k | top_p | min_p | repeat_penalty |
|----------------------------|------|-------|-------|-------|----------------|
| LFM2.5-1.2B-Instruct       | 0.1  | 50    | —     | —     | 1.05           |
| LFM2.5-1.2B-Thinking       | 0.1  | 50    | 0.1   | —     | 1.05           |
| LFM2 text + LFM2.5-JP      | 0.3  | —     | —     | 0.15  | 1.05           |
| All VL models              | 0.1  | —     | —     | 0.15  | 1.05           |

Changes

• Snippet system — Added samplingFlags/samplingParams props to all 6 snippet components so each model page passes its correct params
• 17 model pages — Added per-model sampling props to TextTransformers, TextLlamacpp, TextVllm, VlLlamacpp
• 6 deployment guides — Fixed sampling params in llama.cpp, Transformers, vLLM, Ollama, LM Studio, MLX; used extra_body for non-standard OpenAI client params; fixed
  engine-specific field names (repeat_penalty for Ollama/LM Studio, repetition_penalty for vLLM/llama-server/MLX)
• 2 notebooks — Fixed LFM2_Inference_with_llama_cpp.ipynb and quickstart_snippets.ipynb with correct params