#2306 web app: the default temperature is 0.6; the Qwen3.8 model card and `generation_config.json` say 1.0

Beschreibung

`serve/web/app.js` starts a chat with `temperature: 0.6, top_p: 0.95, top_k: 20`
(`const DEFAULTS = {thinking: "high", temperature: 0.6, ...}`). 0.6 was the earlier Qwen3 recommendation for thinking
mode. For Qwen3.8-Flash-Next the model card's Best Practices say thinking mode `temperature=1.0, top_p=0.95, top_k=20,
min_p=0`, instruct (non-thinking) mode `temperature=0.7, top_p=0.80, top_k=20, presence_penalty=1.5`, and the repository's
`generation_config.json` has `temperature 1.0, top_p 0.95, top_k 20`. DETAILS.md measures sampled decoding with the same
1.0 / 0.95 / 20.

Suggestion: default the web app to 1.0 (thinking mode), or take the defaults from the run config's `sampling` block (and
from the model's `generation_config.json` where setup knows it), so the page, the API's configured defaults and the model
card agree. Small, but a user comparing the page's answers with API answers at the card's settings gets different
sampling without noticing.

Not checked: the Coder / Swift variants' cards (their recommendations may differ from the base model's).