Add per-vault user instructions for LLM context

Add configurable chunk size with import-time selection

Rename generation phase to Preparing response

Add generation phase indicators and stop button

Add context window setting, copy button, similarity indicators, expandable sources

fix: Fix slider truncation jumping

feat: Add Context Window setting

docs: Add detailed README

Fix GPU backend and OpenCL access for on-device LLM

Add STATUS.md with current development state

Fix LLM model loading via SAF content URI

Fix OOM crash: memory-map embedding model instead of readBytes

Replace DJL HuggingFace tokenizer with pure-Kotlin SentencePiece

Fix SAF model loading crashes and missing embedding load in import

Replace placeholder tokenizer with DJL HuggingFace tokenizer