v0.5.11

Added

• boundary CJK tokenizer backend — splits Japanese text at script boundaries (kanji/hiragana/katakana transitions) instead of pure n-gram expansion, producing ~3-5x fewer tokens while maintaining search recall. Now the default backend when sudachipy is not installed
• fix parameter for memory_health_check — when True, automatically re-indexes stale/unindexed notes and removes orphan index entries
• DETAILED_EXCLUDE_FIELDS constant — auto_keywords and work_log_keywords are now excluded from mode="detailed" search results (kept in mode="debug")

Changed

• memory_search docstring now documents mode→behavior mapping (quick/detailed/debug presets and their effect on compact, feedback expansion, and explain settings)
• memory_state_add docstring now includes replace parameter usage example and type clarification (replace must be a list, not a single string)
• memory_evidence error message when both paths and task_id are omitted now suggests correct usage
• memory_health_check annotation changed from READONLY to IDEMPOTENT to reflect the new fix parameter