Add schema v2 with stable IDs and migration

[polymorcodeus]

Here's the same work rewritten to match the lnk #23 voice (imperative title, ## Summary opener, bold-lead-in bullets, a command table, guardrails section, make check closer):

Summary

Adds schema v2: stable IDs and created_at/updated_at on every entity (Shelf, Collection, Mark), plus an explicit book migrate command to upgrade existing v1 shelf files in place. Completes the schema-v2 spine ahead of the SQLite index work and the v2.0.0 release.

What's new

• Stable IDs - shelves and collections get deterministic-seed IDs (marks already ship catalog_id), decoupling identity from the human-readable name so renames stop being destructive.
• Timestamps - created_at/updated_at on all entities, plus deleted_at on Mark (schema only; soft-delete behavior lands separately).
• book migrate - non-interactive v1-to-v2 upgrade, gated behind --confirm like other writes.

Usage

Command	Purpose
book migrate	upgrade all v1 shelf files in shelf.d/ to v2 schema

Behavior changes / guardrails

• Migrated rows get empty timestamps, so two machines migrating the same v1 file produce byte-identical output; only entities created after v2 carry real timestamps.
• book migrate is idempotent: re-running on v2 files is a no-op and reports "N upgraded, M already v2".
• Git-first rollback: inside a git repo, migration requires a clean tree and writes no backups; outside git, it keeps a one-time .bak per file.
• IDs are seeds only: once assigned they are stored and never recomputed, so a later rename keeps identity.

Testing

• Golden v1-to-v2 fixture migration, idempotency (second run no-op), git-clean/dirty, and .bak fallback coverage.
• Migrated files pass round-trip idempotency.
• make check runs fmt, vet, lint, and tests; all green.