MiseLedger 0.3.1

Fixed

• Multi-term search no longer pegs a CPU for minutes on large archives: FTS ranking runs first in a bounded, materialized candidate pool, the relations boost applies only to that pool, and relations gained source/target-item indexes. A query that previously never returned on a 1.4M-item archive now answers in seconds (#9).
• Native imports and crawl sessions skip files whose size and mtime match the scan manifest (content-hash fallback when only mtime differs), report files_parsed/files_skipped, and persist each file’s scan row as soon as its records are committed, so interrupted catch-up runs on large archives make durable progress instead of restarting from zero. --since and --full bypass the fast path; dry runs record nothing (#10).
• The OpenCode adapter skips session_diff JSON arrays instead of emitting a parse warning for every file (#11).