This is phase 7 of this proposal: PyRIT Unified Registry — Design & Rationale

Adds normalized persistence for target, scorer, converter, scenario, seed, and attack identifiers.

• Stores identifiers once by content hash, with ordered relationships for nested component graphs.
• Adds foreign-key links from conversations, prompts, scores, scenario results, and attack results.
• Retains existing identifier JSON as the read source while dual-writing normalized rows for backward compatibility.
• Includes a best-effort Alembic backfill that leaves links nullable when legacy data is malformed.
• Centralizes graph persistence in MemoryInterface so SQLite and Azure SQL share the same behavior.
• Adds coverage for deduplication, graph relationships, dual writes, and migration backfills.