Rewrite technical architecture doc to match current app (Cielo)

Make Medium the default deep tier; pin planning to Sonnet

Add uniform Light/Deep model tiers per dialogue + Deep/Medium/Light switch

docs: consolidate all model-tier benchmarking into one reference doc

docs: add internal architecture memory reference

review structure for structured dialogue

wire the character voice to the system with py file

debug chat

litellm migration + cross-character exclusion management

create philosophic thread

start philosophical explanation logic

remove bugs for short responses to exit the conversation

study on model update