Fix benchmark timing and minor warnings

[HFTrader]

Summary

• Fix timing bracket: Add a thread barrier before start_time in universal_benchmark so that thread creation overhead is excluded from throughput measurement. Previously, start_time was captured before spawning threads, inflating measured wall time by 10-50μs per thread.
• Initialize variables: Initialize res1/res2 in cuckoo_insert to silence -Wmaybe-uninitialized. Both are always set by try_find_insert_bucket() before use, but the compiler cannot prove this statically.
• Fix unused parameter: Remove unused parameter name in upsert_fn lambda to silence -Wunused-parameter.

Test plan

• All existing unit tests pass
• Universal benchmark builds and runs correctly
• Benchmark results are consistent with pre-change (slightly higher throughput due to excluded thread creation overhead)