Multilingual text detection and routing accuracy, specifically for non-Latin scripts (Armenian). Core functionality failure in language analysis and routing.

Armenian text is classified as having no letters. At `6a5819129eb220570792e417e49723d697efd76f`, `analyse("Հայերեն")` returns:

```python
{'script': 'unknown', 'script_profile': {}, 'language': None,
'is_english': True, 'non_latin_fraction': 0.0}
```

The characters pass `isalpha()`, but `_SCRIPT_RANGES` has no Armenian entry, so both `detect_script` and `script_profile` discard them. `Router.route` treats `unknown` as “no letters detected” and selects the default checkpoint (English by default). I'd expect Armenian text to select the multilingual route, while genuinely letter-free input such as `12345` keeps the existing fallback.

Reproduced on Windows with Python 3.13 by loading the dependency-free language module directly from the checkout:

```python
import importlib.util

spec = importlib.util.spec_from_file_location("laya_lang", "laya/lang.py")
lang = importlib.util.module_from_spec(spec)
spec.loader.exec_module(lang)
print(lang.analyse("\u0540\u0561\u0575\u0565\u0580\u0565\u0576")) # Հայերեն
```

The English control (`Please refund the charge`) returns `latin` / `is_english=True`; the numeric control returns `unknown`. This checks the detector only; the routing consequence above is from reading `Router.route`. No model weights were loaded and no inference accuracy was measured.

AI assistance: Codex investigated this and ran the reproduction.