perf(java): pre-compile WASM module once and reuse across instances

[nicklasl]

Summary

• Pre-compile the WASM module once in SwapWasmResolverApi constructor using MachineFactoryCompiler.compile(WasmModule) and reuse the compiled machine factory across all WasmResolveApi instances.
• Previously, every updateStateAndFlushLogs call created a new WasmResolveApi, triggering full AOT WASM-to-JVM bytecode compilation each time via MachineFactoryCompiler::compile(Instance).

Test plan

• mvn compile passes
• Verified all test failures are pre-existing (missing deps, missing env vars)
• Verify in staging that state updates no longer cause CPU spikes from WASM recompilation

🤖 Generated with Claude Code

[nicklasl]

Benchmark Results

Ran a 2-minute CPU benchmark comparing this branch against main, simulating production load: ~1 flag evaluation every 60ms (~2000 total) with state updates every 30s (3 updates total). CPU time measured via OperatingSystemMXBean.getProcessCpuTime().

Metric	This branch (optimized)	main (baseline)
CPU time	3.94s	8.41s
Evaluations	2001	2000
State updates	3	3
CPU/eval	1.97ms	4.21ms

~53% CPU reduction — the savings come from eliminating MachineFactoryCompiler.compile() on each state update, since the compiled WASM module is now reused across instances.

How the benchmark works

A JUnit test creates a SwapWasmResolverApi with a real resolver state and uses a ScheduledExecutorService to run flag evaluations every 60ms and trigger updateStateAndFlushLogs every 30s, measuring process CPU time over 2 minutes. A shell script automates running the benchmark on both branches and prints a side-by-side comparison.

[andreas-karlsson]

Actually feels like this should be static for this to be any speedup?

I changed my mind.