The truncateAtWord function was slicing strings by byte position instead of
character position. When truncating text with multi-byte UTF-8 characters
(like CJK), this could cut in the middle of a character, creating invalid
UTF-8 and causing gRPC marshaling errors.

Fixed by converting to runes before truncation to ensure we always cut at
proper character boundaries. Added test cases for CJK characters.

Fixes #5276

🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>