Benchmark Brev against ReAI Peppol generator

[gregjotau]

Summary

• benchmark Brev against the Digipost Peppol generator currently used by ReAI
• use preconstructed invoices with the same business values and returned byte-array output
• verify both outputs contain equivalent values and pass the current Peppol Billing 3.0.21 PHIVE rules
• document throughput, allocation, artifact, and class-count results

JDK 26 results

• Brev: 263,003 ops/s; 17,568 B/op
• Digipost: 38,267 ops/s; 272,016 B/op
• Brev: 6.87x throughput and 15.48x less allocation
• runtime envelope: 88,595 bytes / 45 classes vs 233,648 bytes / 192 classes

Verification

• ./gradlew clean build
• JMH: 3 x 1s warmup, 5 x 1s measurement, 2 forks, GC profiler

This establishes a continuation gate only. ReAI should not migrate until shadow comparison covers production-shaped invoices, credit notes, discounts, VAT combinations, rounding, and attachments. Not merged; ready for manual review.