fix: TTS ack-bank served stale voice after voice-config changes; add website/English pronunciation rules

RuntimeConfiguredTTSProvider (the live 'dynamic' TTS provider, resolves
voice from DB-backed config) never overrode cache_fingerprint(), so it
fell back to the base class's empty string. PrebakedAckBank keys its
on-disk cache on that fingerprint, so short filler phrases like
'Секунду' kept serving audio baked with the previous ElevenLabs voice
even after a voice change, while full LLM replies (cached inside the
resolved provider itself, keyed on its own voice id) already used the
new voice — explaining why callers heard two different voices in the
same call. Fix: delegate cache_fingerprint() to the resolved provider.

Also extend the voice delivery_hint so the model transliterates website
addresses and English words/abbreviations into spoken Cyrillic instead
of leaving raw Latin text for the TTS engine to mangle (egov.kz was
coming out as 'эговкз').

services/ai_orchestrator_service/operator_persona.py

@@ -143,7 +143,12 @@ def operator_system_prompt(*, language: str, channel_label: str, is_voice: bool,
         "Short hotline or service numbers (e.g. 1414, 109) must be spelled out the way people say them as a code, "
         "grouped and read naturally (\"1414\" as \"четырнадцать четырнадцать\", not \"тысяча четыреста четырнадцать\"). "
         "Calendar dates must use the correct spoken grammatical case (\"25 числа\" as \"двадцать пятого числа\", "
-        "not \"двадцать пять число\"; \"14 марта\" as \"четырнадцатого марта\")."
+        "not \"двадцать пять число\"; \"14 марта\" as \"четырнадцатого марта\"). "
+        "Never leave a website address, domain, or English word/abbreviation in raw Latin script — the TTS engine "
+        "slurs it into gibberish (e.g. \"egov.kz\" comes out as \"эговкз\"). Transliterate it into how a person "
+        "actually pronounces it aloud, with an explicit pause word for punctuation: write \"egov.kz\" as "
+        "\"игов точка кэ-зэт\", write \".kz\"/\".com\" as \"точка кэ-зэт\"/\"точка ком\", spell out an acronym or "
+        "English word phonetically in Cyrillic (\"IT\" as \"ай-ти\", \"email\" as \"имейл\")."
         if is_voice
         else "The reply should read like a concise message from a live first-line operator."
     )

services/ai_voice_runtime_service/runtime_tts_provider.py

@@ -114,6 +114,10 @@ class RuntimeConfiguredTTSProvider(TTSProvider):
                 self._provider_cache[cache_key] = provider
         return provider
 
+    def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
+        provider = self._provider_for_language(language)
+        return f"{provider.name}:{provider.cache_fingerprint(language, style_hints=style_hints)}"
+
     def synthesize(self, text: str, *, language: str | None = None, style_hints: dict[str, object] | None = None):
         provider = self._provider_for_language(language)
         return provider.synthesize(text, language=language, style_hints=style_hints)