Durable archive recovery, llama.cpp performance profiles for qwen

Add durable archive recovery and llama.cpp runtime performance tuning
Resolves data race conditions between thread rename/append/delete and
background embedding, replaces delete+add cycles with atomic LanceDB merges,
adds durable recovery with exponential backoff, and introduces hardware/GGUF-guided
performance tuning for llama.cpp. Promotes conversation generation to Beta.

1.1.0

What's Changed

• Feat/model catalog and llama release fix by @ChrystianSchutz in #14

Full Changelog: 1.0.0...1.1.0

ThreadShelf v1.0 - Initial public release

ThreadShelf v1.0 - Initial public release