feat: add inner loop and compile-once-run-many optimization for Java benchmarking

[misrasaurabh1]

Summary

• Add inner loop in Java test instrumentation for JIT warmup within single JVM
• Implement compile-once-run-many: compile tests once with Maven, then run directly via JUnit Console Launcher (~500ms vs ~5-10s per invocation)
• Add fallback to Maven-based execution when direct execution fails
• Update parsing to handle JUnit Console Launcher output format

Changes

Inner Loop Instrumentation

• Modified _add_timing_instrumentation() to wrap test code in a for loop
• Reads CODEFLASH_INNER_ITERATIONS from environment (default: 100)
• Each iteration prints its own timing markers with unique iteration ID

Compile-Once-Run-Many Optimization

• New _compile_tests() - compiles tests once with mvn test-compile
• New _get_test_classpath() - gets classpath from Maven
• New _run_tests_direct() - runs tests directly via JUnit Console Launcher
• Reduced overhead from ~5-10s to ~500ms per JVM invocation

Parsing Updates

• Added fallback to use subprocess stdout when XML system-out has no timing markers
• This handles JUnit Console Launcher output format correctly

Benchmark Flow

Python Process
│
├─ Step 1: mvn test-compile (one-time: ~5-10s)
├─ Step 2: mvn dependency:build-classpath (one-time: ~2-3s)
│
└─ Step 3: For each outer loop (1-3 JVM invocations):
      │
      └─ java -cp ... ConsoleLauncher ... (~500ms overhead)
           │
           └─ Inner Loop (100 iterations within JVM):
                ├─ Iteration 0: print timing marker, run test, print result
                ├─ Iteration 1: print timing marker, run test, print result
                ├─ ...
                └─ Iteration 99: print timing marker, run test, print result

Test plan

• All 32 Java instrumentation tests pass
• E2E tests for inner loop timing markers
• E2E tests for multiple test methods with inner loop
• Fallback to Maven execution when direct execution fails

🤖 Generated with Claude Code