PromptDesk model ladder

VERIFIED Redis-ranked cheapest models; BullMQ attempt maps to rank index.

Setting	Default
Provider order	gemini,openai
Rank refresh	12h
Top N	3
Job attempts	CHAT_JOB_ATTEMPTS=3
LLM timeout	30s

Attempt N uses model index N-1 on the current provider; after exhausting attempts, failover to the next provider in order.