fix: repair legacy desktop upgrade migration and replay crashes

[paul-bouzian]

Why

• Desktop upgrades can fail when an existing effect_sql_migrations table already contains different migration names for ids 17 through 22 from an older release line.
• In that state, the current thread metadata migrations are skipped, 023_ProjectionThreadsAssociatedWorktreeRef still runs, and startup crashes because associated_worktree_branch does not exist.
• Some legacy desktop databases also contain thread.archived or thread.unarchived orchestration events, which the current release cannot decode during replay, causing the backend to restart in a loop even after the schema issue is repaired.

How

• Make migration 23 repair the full current thread/message metadata shape before backfilling associated_worktree_ref.
• Detect missing columns from the current 17-23 migration line and add only the ones that are absent, so healthy databases keep their existing data untouched.
• Add migration 24 to normalize legacy archived/unarchived thread events into decodable thread.meta-updated rows that preserve threadId and updatedAt.
• Add regression tests for both the migration-ledger mismatch and the legacy archived-event replay path.

Tests

• bunx vitest run apps/server/src/persistence/Migrations/023_ProjectionThreadsAssociatedWorktreeRef.test.ts apps/server/src/persistence/Migrations/024_NormalizeLegacyArchivedThreadEvents.test.ts
• bun fmt
• bun lint (passes with existing repo warnings only)
• bun typecheck

Review Setup

• No extra setup is required beyond the normal app deploy.
• To manually verify the upgrade path, use a desktop profile whose effect_sql_migrations table already contains the legacy 17-22 rows and whose orchestration_events table may include legacy thread.archived rows, then launch the app and confirm startup no longer crashes during replay.