Add soft delete, gc, and doctor for bookmark catalogs

[polymorcodeus]

Summary

Adds soft delete, garbage collection, and merge-doctor tooling to bookmarks, so a removal is reversible and post-merge duplicates are detected and auto-resolved. Completes the catalog-maintenance phase of the v2 hardening.

What's new

• Soft delete - mark remove stamps deleted_at instead of slicing the mark out, so a removal stays recoverable until garbage collection.
• Trash + restore - mark ls --trash lists soft-deleted marks by their stable catalog_id; mark restore --id (or --shelf/--collection/--url) clears deleted_at.
• Garbage collection - book gc purges marks soft-deleted past the retention window (30 days by default) behind a required --confirm.
• Merge doctor - book doctor (alias book sync) reports duplicate marks, same-ID different-content conflicts, schema-v1 files, stale index entries, and stray .tmp/.bak debris.
• Auto-merge - book doctor --fix dedups identical merge artifacts and surfaces true conflicts for manual resolution.

Usage

Command	Purpose
mark ls --trash	list soft-deleted marks
mark restore --id <catalog_id>	restore a trashed mark
book gc [--retention-days 30]	purge expired soft-deleted marks
book doctor / book sync	report catalog health
book doctor --fix --confirm	auto-merge duplicate marks

Behavior changes / guardrails

• mark remove now soft-deletes instead of removing: the mark keeps deleted_at and is hidden from list/get/search until book gc purges it.
• book gc and book doctor --fix both require --confirm because they rewrite shelf files.
• Duplicate classification is content-addressed: identical title, tags, and deleted state auto-merge; any difference becomes a manual conflict.
• book doctor --fix keeps the first occurrence of an identical duplicate and skips index reconciliation when true conflicts remain, since the index catalog_id primary key cannot represent duplicate IDs.
• book doctor flags schema-v1 files (pointing at book migrate) and stale index entries (pointing at book index sync).

Testing

• Unit tests cover duplicate detection and conflict classification, auto-merge resolution, schema-v1 detection, stray debris, and index staleness.
• make check runs fmt, vet, lint, and tests; all green.