#3142 Dashboard: show the loaded model in the top bar and on the Model state card

本文

Screenshots of the Monitor never say which model is running. This shows it in the top bar (every tab) and next to the
Model state title:

`Qwen3.8-Flash-Next · Swift 1.5 IQ3_S`

## Changes (26 lines, opt-in)

- **`serve/server.py`:** an optional config key `"display_name"`. `/metrics` reports `engine.display_name` and
  `engine.mode`, a free-form note an engine adapter can set on the Service (`svc.mode_note`, `None` here); `/health`
  reports `display_name`. Nothing changes when neither is set: the label then shows the model id.
- **`serve/web/`:** `renderModelLabel()` puts `<display_name or model id> · <mode>` in the header and on the Model state
  card (muted text, ellipsized in the header).

Xolotl uses the mode note for DeepSeek-V4.1's routing mode (Indras-Mirror/Xolotl#13), e.g.
`DeepSeek-V4.1-Flash-1017 · dsx 2-bit · light routing (0.25 / 0.15)`. Here it is only the hook.

## Testing

- Mock engine (`serve/server.py --engine mock`) with a `display_name` in the config: `/health` and `/metrics` carry it,
  and the Monitor renders the label (headless Chrome).
- `python -m unittest serve.test_server`: 337 OK (3 skipped). `serve.test_monitor`: 10 OK.
- On Windows 11 with an RTX 3090; nothing platform-specific.