[tx] Add separate JIT vs post-JIT timing measurement to memory benchmark

[raulchen]

Summary

• Separate JIT compilation time from post-JIT runtime measurement by sending multiple requests per test
• First request triggers JIT compilation and is measured separately
• Subsequent requests (default: 3) measure actual post-JIT performance and are averaged
• Extract common timing logic into _run_timed_requests helper method
• Use random tokens (with fixed seed) for reproducible benchmarks

Changes

• Add jit_e2e_sec field to capture first request (JIT) time
• Rename client_e2e_sec to post_jit_e2e_sec for clarity
• Add --num-measurement-iters flag (default: 3) for post-JIT iterations
• Update CSV output and summary tables to show both metrics

[gemini-code-assist]

Code Review

The pull request successfully introduces separate timing measurements for JIT compilation and post-JIT runtime, providing a clearer understanding of performance characteristics. The changes involve updating the BenchmarkConfig and TestResult data structures, modifying the test execution logic to perform warmup and multiple measurement iterations, and updating output formats (CLI summary and CSV) to reflect the new metrics. The addition of the --num-measurement-iters CLI flag makes the benchmark more configurable.

[raulchen]

/gemini review

[gemini-code-assist]

Code Review

The pull request successfully introduces separate timing measurements for JIT compilation and post-JIT runtime, significantly enhancing the memory benchmark's insights. The changes are well-integrated across the configuration, data structures, execution logic, and reporting. The addition of _run_timed_requests effectively encapsulates the new measurement methodology, and the updated output formats provide clear visibility into the new metrics.