fix: remove deterministic count response intercept

[adgk2349]

Summary

• remove the hard-coded count/list response interceptor from general chat
• route count-based requests through the normal contextual inference flow

Root Cause

Requests such as "소고기 굽는 법을 핵심 3가지만 정리해줘" matched a numeric-count heuristic before model inference. The fallback template ignored the requested topic and returned generic task items.

Impact

The local model now receives count-based requests together with conversation context, so it can answer the actual topic instead of a static template.

Validation

• ./.tmp-mm-venv/bin/python -m pytest sidecar/tests/test_local_inference_sanitize_conversation.py -q
• ./.tmp-mm-venv/bin/python -m py_compile sidecar/local_ai_core/reasoning/strategies/general_chat_sections/general_chat_execution_mixin.py
• Manual local-session simulation confirmed conversational memory injection across a multi-turn stew discussion.