- Remove duplicate function definitions with hardcoded "AI-оператор" strings
(ai_voice_runtime, ai_orchestrator, voice_name_config, voice.py)
- Remove unreachable dead code after return in ai_voice_runtime
- Add SQL LIMIT to 17 unbounded queries across 12 services to prevent OOM
- Move Python-side filtering to SQL WHERE in reporting_service
- Downgrade 19 logger.warning to logger.info for normal-flow events in media_runtime
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When a new voice session starts and the customer already has a trusted
display_name on file (e.g. from a previous call), automatically set
name_status to 'name_obtained' so the AI greets them by name instead
of asking again.
- Block old _voice_downstream_name_update from overwriting name_obtained status
unless user explicitly says 'меня зовут X' (explicit_candidate only)
- Move _persist_voice_name_state to run AFTER LLM extraction, not before,
so LLM always gets a chance to override the regex result
- Fixes bug where phrases like 'можешь назвать их' were incorrectly
captured as customer name, overwriting the real name
Yera All