Add formal performance benchmarks

[walterra]

Overview

PERFORMANCE.md currently references benchmark numbers without actual test data. We need to run formal benchmarks and document real results.

Benchmark Scenarios

1. File Ingestion

   • Small file: 10,000 docs (~1 KB each, ~10 MB file)
   • Large file: 1,000,000 docs (~1 KB each, ~1 GB file)
   • Measure: time, throughput (docs/sec), memory usage

2. Reindexing

   • Local reindex: 10,000 docs with simple transform
   • Measure: time, throughput, memory

3. Cross-Version Reindex

   • ES 8.x → ES 9.x: 10,000 docs
   • Measure: time, throughput, memory, overhead vs same-version

4. Transform Complexity

   • No transform vs simple vs complex transform
   • Impact on throughput

5. Buffer Size Impact

   • Test different bufferSize values (1 MB, 5 MB, 10 MB, 20 MB)
   • Document optimal settings for different document sizes

Test Environment

Document hardware specs:

• CPU, RAM, disk type (SSD vs HDD)
• Elasticsearch version
• Node.js version
• Local vs remote ES

Deliverables

• Benchmark script in scripts/benchmark.js
• Update PERFORMANCE.md with real results
• Add benchmark results to CI (optional)
• Document reproducibility steps

Acceptance Criteria

• Real benchmark data replaces placeholder content in PERFORMANCE.md
• Results are reproducible with documented steps
• Multiple scenarios covered (file sizes, reindex, transforms)
• Memory usage verified to be constant