[stonecrop] optimize field trigger rollback memory usage

[Alchez]

Description

The current snapshot-based rollback implementation creates a full deep copy of the entire record before executing field trigger actions, which may introduce memory overhead for:

• Large documents (100KB+ records)
• High-frequency field changes
• Rapid sequential edits

Tasks

• Measure baseline memory usage with realistic record sizes
• Profile memory allocation patterns during field trigger execution
• Benchmark performance with high-frequency changes (e.g., 100 changes/sec)
• Evaluate alternative snapshot strategies:
  • Structural sharing
  • Diff-based snapshots
  • Copy-on-write
• Implement optimizations if measurements show significant overhead
• Add performance tests to regression suite

Success Criteria

• Memory usage profiled and documented
• Performance benchmarks established
• Optimization implemented if >10% overhead detected
• No breaking changes to public API