refactor: per-lib database artifacts with consolidate pipeline

[laradji]

Summary

• Per-lib artifact databases: The scraper now writes one .db file per library into ./artifacts/ instead of inserting directly into the main database. Each artifact carries its own lib_id in the meta table and is self-contained.
• New cmd/consolidate command: Merges all per-lib artifact .db files from ./artifacts/ into the main deadzone.db in a single atomic transaction. Idempotent — replaces existing rows per lib_id with a full delete-then-insert within one tx.
• OpenArtifact in internal/db: New entry point for opening per-lib databases with lib_id identity tracking, validation, and mismatch detection (ErrArtifactLibIDMissing, ErrArtifactLibIDMismatch).
• Schema version bump to v2: Adds schema_version to the meta table and cross-checks on open, surfacing ErrSchemaMismatch for incompatible databases.
• Justfile recipes: Added scrape, consolidate, serve, and clean recipes for the new two-stage pipeline.

Motivation

Decouples per-lib scraping from the shared database so individual libraries can be re-scraped independently (just scrape lib=/org/project) without touching other libs' data. The consolidate step is explicit and atomic — partial failures leave the main DB untouched.

Test plan

• just test — new consolidate_test.go covers validation pass (embedder mismatch, schema mismatch, missing lib_id), merge pass (single/multi artifact, idempotent re-merge), and rollback on mid-merge failure
• just build — all three commands compile
• just scrape lib=/some/lib && just consolidate — end-to-end artifact creation and merge

Fixes #28