storage: migrate artifact metadata to SQLite

[likun666661]

Summary

• add the artifact schema and metadata repository to the shared runtime.sqlite operational authority
• cut over legacy artifacts/metadata.jsonl once under the artifact writer lock, with fingerprinted journal validation and fail-closed old-writer detection
• keep payload bytes as files while reusing the durable staging/link and purge-intent recovery protocol against SQLite metadata transactions
• wire CLI, Desktop, Runtime Host, and Headless artifact writers to the SQLite-backed store and close their database leases
• teach Harbor trajectory image collection to read artifact_records from runtime.sqlite, with legacy JSONL fallback

Crash and migration guarantees

• interrupted legacy import rolls back and resumes idempotently
• payload linked without a committed metadata row is removed during recovery
• committed metadata preserves the payload and removes publication staging residue
• interrupted purge completes payload and SQLite metadata cleanup from the durable purge intent
• a changed legacy JSONL source after completed cutover fails closed
• a new workspace does not create metadata.jsonl

Verification

• npm --workspace @maka/storage test — 849 passed, 1 skipped
• npm --workspace @maka/runtime-host test — 414 passed
• Storage, Runtime Host, Headless, CLI typechecks
• Desktop main build
• Biome checks and Python py_compile
• direct SQLite trajectory metadata smoke

The full local Harbor adapter file also encountered two unrelated environment-sensitive Python 3.9 failures in the Kimi/Codex adapter tests; the modified trajectory path was validated separately.

Part of #1649. The selected-session bundle / backup relational-closure work remains the next planned PR (Stage 5).