fix(#34): split long Telegram messages at natural boundaries

[arniesaha]

Summary

Fixes #34 — long responses were getting cut off mid-sentence with a ▍ cursor artifact.

Changes

• findSplitPoint(text, maxLen) — new helper that finds the best split point using priority order: paragraph (\n\n) > newline (\n) > sentence end (. , ! , ? ) > word boundary (space) > hard cut
• Streaming split loop — now uses findSplitPoint() instead of inline lastIndexOf("\n") logic
• Continuation messages — now sent via sendWithFormat() (HTML-formatted) instead of raw sendMessage(), plus a 200ms delay before sending to preserve ordering
• splitMessage() — updated to use findSplitPoint() and trim leading whitespace from remainder chunks

Acceptance Criteria

• ✅ Responses longer than 4096 chars split into multiple sequential messages
• ✅ Splits happen at natural boundaries (paragraph > sentence > word)
• ✅ No cursor artifact or abrupt mid-sentence cutoff
• ✅ Each chunk sent in order with short delay

Testing

Ran unit tests validating: short messages pass through, paragraph splits, sentence splits, no cursor artifacts in chunks, all chunks ≤ 4000 chars.