FIX Stop using deprecated AttackResult.attack_identifier in attack_service

[romanlutz]

Summary

Migrate attack_service.py away from the deprecated AttackResult.attack_identifier property (scheduled for removal in 0.15.0).

Changes

Production code (pyrit/backend/services/attack_service.py):

• Replace ar.attack_identifier with ar.get_attack_strategy_identifier() (the non-deprecated equivalent)
• Stop writing the deprecated attack_identifier key in update_fields during converter merging — only atomic_attack_identifier (the source of truth) is now updated

Tests (tests/unit/backend/test_attack_service.py, tests/unit/memory/test_memory_models.py):

• Update existing assertions to use get_attack_strategy_identifier() and check atomic_attack_identifier in update_fields
• Add test for flat/legacy atomic identifier fallback converter merge path
• Add assertion that deprecated attack_identifier column is not written to update_fields
• Add test proving DB read prefers atomic_attack_identifier over stale attack_identifier

Why this is safe

• AttackResult.attack_identifier just delegates to get_attack_strategy_identifier() — identical return value
• All attacks (GUI-created and scenario-based) always set atomic_attack_identifier
• Legacy DB rows without atomic_attack_identifier get it reconstructed on read via AttackResultEntry.get_attack_result()
• The attack_identifier DB column becoming stale is harmless: get_attack_result() always prefers atomic_attack_identifier

Eliminates ~42 DeprecationWarning emissions during unit tests.