Roster slugs through Home Assistant's real slugify (P2 fix)

The per-child entity ids are derived from the child name by Home
Assistant's real slugifier, which transliterates every script through
unidecode; the hand-rolled Latin-only mirror in sensor.py dropped
non-Latin characters instead (slugging 京子 to "unknown"), so the
roster's slugs diverged from the actual entity ids and the
party-board/strategy auto-discovery silently failed for non-Latin
households.

Use homeassistant.util.slugify at runtime, composing the registry's
64-character entity-id cap the roster needs (verified against HA
2024.6: slugify itself does not truncate; the entity registry
truncates the full <domain>.<slug> id). Give the mock harness a
faithful homeassistant.util.slugify stub — python-slugify's frozen
legacy pipeline with unidecode, verified byte-for-byte against real
python-slugify across Latin, Cyrillic, CJK and Greek name shapes —
and cover the non-Latin resolution contract in tests (roster slug ==
the object id HA's entity-id derivation produces).

Feature-review finding (P2): auto-discovery slug must match Home
Assistant's real slugify.

pyproject.toml

@@ -21,6 +21,7 @@ dev = [
     "pytest>=8.0.0",
     "pytest-asyncio>=0.24",
     "pyyaml>=6",
+    "unidecode>=1.4.0",
     "uvicorn>=0.30",
     "voluptuous>=0.14",
 ]