Add reader core scaling measurements with hot and cold caches

[tom-clickhouse]

This adds reproducible reader CPU scaling measurements for Blog 6 11c, using the original D1/D2 drill-down queries against the fixed 113.2B-quote dataset. The current cache-controlled sweep measures 2, 4, 8, 16, 30 and 59 cores on one fixed 59-vCPU / 236-GiB node, with max_threads controlling query parallelism. It does not compare six physically resized services.

The selected visual compares Hot caches with Cold data caches and query-condition caching off. Pooled uncompressed throughput rises from 0.803 to 15.560 GB/s for Hot and 0.731 to 9.848 GB/s for Cold. Gains taper at larger settings; the package does not claim perfect linear scaling. A separate condition-cache-on cold treatment and earlier 16/59-vCPU exploratory ledgers remain intact, without being spliced into the selected curves.

The cache pilot demonstrated that filesystem-cache-off alone still read through the userspace page cache. Cold treatments therefore bypass filesystem, page, and distributed data-cache reads per query. Profile counters verify positive S3 reads and zero page/distributed reads for Cold, and page hits with zero S3 reads/misses for Hot. Query-result caching remains disabled. Metadata/mark caches are not globally purged.

Included: exact source/treatment SQL, shuffled trial plans, append-only HTTP records, full query-log CSVs/ProfileEvents/Settings, capacity/version/data/cache checks, derived statistics, reports, offline analysis, live repeat instructions, and SHA-256 manifests. derive_visual.py reproduces the exact 11c payload and resource CSV: CPU used/headroom relative to the selected core budget and average measured per-query memory peaks. Throughput per core uses consumed CPU seconds, matching 11a.

Validation:

• 432 cache-controlled measured queries plus 24 pilot/warmup executions; all 456 server records collected.
• Zero failures, missing logs, result-cache reads or removed outliers; unchanged active data parts and service identity/capacity/version.
• Offline cache analysis and resource derivation pass; exact 11c JSON matches the production visual input.
• All 526 cache-package file hashes verify; private-secret scan, README links, and staged patch checks pass.

Start at full-path-realtime/quotes/clickhouse-cloud/experiments/reader-thread-scaling/README.md, then 2026-10-11-cache-modes/README.md for selected cases and reproduction steps. Original Run 2 continuous-ingestion evidence remains separate in PR #44; this experiment does not recreate its historical cache state or 3.25 GB/s display endpoint.