perf: optimize XLIFF import with DBAL bulk operations

Replace individual persistAll() calls with batched DBAL operations
achieving 6-33x performance improvement depending on environment.

Performance improvements (DDEV/WSL2):
- 1MB file (4,192 records): 23.0s → 3.7s (6.18x faster)
- 10MB file (41,941 records): 210.4s → 8.7s (24.18x faster)
- Performance scales logarithmically with dataset size

Implementation (5-phase architecture):
1. Validation & pre-processing: Extract unique components/types
2. Reference data: Find/create Environment/Component/Type entities
3. Bulk lookup: Single query fetches all existing translations
4. Batch preparation: Categorize into INSERT vs UPDATE arrays
5. DBAL execution: bulkInsert() and batched UPDATE operations

Technical changes:
- Use ConnectionPool and QueryBuilder for SQL injection prevention
- Batch operations by 1000 records for memory efficiency
- Transaction-safe with explicit commit/rollback
- Maintain single persistAll() for reference data only

Optimized environment (native Linux) achieves up to 33x improvement.

BREAKING: Bypasses Extbase ORM hooks (documented in ADR-001)

Author: CybotTM