Add two-phase message handling for PDF files sent without text (common
in Telegram reply-with-attachment flow):
Phase 1 (pre-OCR, 5s wait):
- Send hint message when bare PDF arrives
- Wait up to 5 seconds for follow-up message with OCR keywords
(e.g. "figures"/"図版") to set OCR flags before processing starts
Phase 2 (during OCR):
- Run processMessage/OCR concurrently while draining the message queue
- Buffer same-chat messages as user instructions for post-OCR LLM turn
- Support cancel keywords ("cancel"/"中止"/"キャンセル") to abort OCR
- After OCR + initial LLM response, merge buffered messages into a
follow-up LLM turn so user instructions are processed with context
Refactors llmWorker into llmWorkerNormal/llmWorkerPDF with shared
helpers (sendResponseIfNeeded, resetMessageTool).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
