LLM 0.32 candidate release brings redesigned message storage model

The LLM tool reached release candidate status for version 0.32. The main improvement is a redesigned database schema that better captures prompts and responses from the latest model families. The key architectural shift involves content-addressable hash identifiers for stored messages. This enables deduplication and allows LLM to represent branching message trees for conversations that split into multiple paths. Support is added for three new models: gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. Users should back up their databases before upgrading, as the schema changes are substantial.