#2311 web: default temperature 1.0, as the model card and generation_config.json say

Descrição

Issue: Resolves #2306

## Summary
The web app started a chat at temperature 0.6 (the older Qwen3 thinking-mode value). The Qwen3.8-Flash-Next model card (thinking mode) and its `generation_config.json` say 1.0 / 0.95 / 20, which DETAILS.md also uses for sampled decoding.

## What changed
- `serve/web/app.js`: `DEFAULTS.temperature` 0.6 -> 1.0. top_p and top_k were already 0.95 / 20.
- `serve/test_web_defaults.py`: reads `DEFAULTS` from app.js and checks 1.0 / 0.95 / 20.

## Extra Notes
- What it leaves alone: settings a user already saved (they are kept in the browser's store and still win), the API's defaults, and the run config's `sampling` block. This is not a speed change.
- How I checked: `python3 -m unittest serve.test_web_defaults` fails on main and passes with the change.
- Not checked: the Coder / Swift variants' cards, as the issue notes.