Large flag performance test

[leoromanovsky]

🎟️ Fixes issue
📜 Design Doc: link if applicable

Motivation and Context

We need a test bench to compare improvements for SDK startup time when huge flag payloads are delivered.

Eppo-exp/sdk-test-data#139

The tests metrics collected aren't meaningful on their own; firstly we are running them on powerful M-series laptops or cloud computers, compared with mobile devices. However the delta between implementation will be useful.

I wanted to have 10k flags, which is far more than our largest deployment, to demonstrate a clear separation with new implementations.

Description

Instantiated the SDK with a UFC containing 10k flags and performs 50k evaluations; it is not concerned with correctness just measuring startup time and evaluation throughput.

We will compare this with an implementation based on flatbuffers to demonstrate:

• Faster startup
• No meaningful impact on evaluation throughput

How has this been documented?

How has this been tested?

🚀 Starting JSON Performance Benchmark
🎯 Key Focus: Startup performance (JSON->objects conversion)
📊 Secondary: Evaluation performance across test cases

📦 Data loaded: 53 MB + 12 test cases (10.07ms)

🔄 Benchmarking JSON Mode...
   🏁 Starting JSON->objects conversion...
   ⚡ Startup complete: 2175ms, Memory: +394MB
   🏃 Running evaluation performance benchmark across all flags...
   📊 Testing 10000 flags with 5 subjects each...
   ✅ Evaluations complete: 50000 in 485ms
   📈 Performance: 103123 evals/sec

🏆 PERFORMANCE BENCHMARK RESULTS:
📊 JSON Mode:
   🎯 Startup (KEY METRIC): 2175ms
   ⚡ Evaluation Speed: 103123 evals/sec
   💾 Memory Usage: 465MB
   📊 Total Evaluations: 50000

🎯 JSON Performance Benchmark completed!
   Ready for FlatBuffer comparison when implemented

[leoromanovsky]

not planning on merging this, of course