[DO NOT MERGE] chDB: test query memory guard on two-vCPU machines - #2488

Purpose

Test whether max_threads_min_free_memory_per_thread=1 GiB unnecessarily reduces query parallelism on the two-vCPU ClickBench machines.

The new chdb-rc-no-final-lz4-i1-thread-memory-0 entry keeps the established configuration unchanged:

• chDB 4.4.0 with chdb-core v26.9.0
• auto_statistics_types=''
• LZ4 with adaptive codec selection disabled
• max_insert_threads=1
• no OPTIMIZE FINAL

The only measured-path change is max_threads_min_free_memory_per_thread=0. It is set on each chDB session immediately before the timer starts, so the SET execution itself is excluded from query time.

max_insert_threads_min_free_memory_per_thread is intentionally unchanged: with max_insert_threads=1, the insert-side limiter cannot reduce parallelism further.

The release-candidate-only post-run profiler is omitted because v26.9.0 predates benchmark/load_profile.py. That profiler ran outside all measured windows and did not affect ClickBench results.

Intended comparison

Compare t3a.small and c6a.large with the chdb-rc-no-final-lz4-i1 results from #2463, focusing on Combined, Cold, Hot, and per-query ratios.

Validation

• bash -n for every shell entry point
• python3 -m json.tool for template.json
• python3 validate-results.py (39,546 files validated; only six pre-existing Infobright warnings)
• local query-wrapper smoke test using the official chdb-core v26.9.0 wheel