Full-history query benchmark commands — reads across all data types  #856

TL;DR

#771 was re-scoped to ingestion benchmarks; this issue carries the query half. Rewire the rpc-hack read benches onto the production store readers as a bench-query subcommand next to bench-ingest, reusing #771's CSV sink and reporting format. Together with #771 this completes the original scope (and #730's).

Scope

• The eight read benches from the prototype: {hot,cold} × {ledgers, txpage, txhash, events}.
• Measure the production read paths in internal/fullhistory/storage/stores/ — e.g. eventstore/query.go + cold_reader.go, ledger/cold_reader.go + hot_store.go, txhash/read_assembly.go + cold_reader.go — not the prototype's own reader implementations (same rewiring principle as Full-history benchmark commands — ingest across all data types #771).
• Cold event/txhash benches must measure the complete correct read, including the MPHF post-filter — an index probe alone benchmarks a correctness bug, not a query.
• --query-concurrency sweeps and the grid runner (concurrency × data type matrix).
• Events query-corpus auto-generation: derive realistic contract-ID/topic terms from the ingested data rather than hand-picked constants.
• XDR-views only (the decided path; no views-vs-decode comparison).
• CSV output in Full-history benchmark commands — ingest across all data types #771's schema (stage,n,n_items,total_ns,p50_ns,p90_ns,p99_ns,max_ns) so Run full-history benchmark commands in CI to catch perf regressions #775 can consume ingest and query reports uniformly.
• Land on feature/full-history at mergeable quality.

Source

rpc-hack scripts/bench-fullhistory/: bench_{cold,hot}_*.go, bench_grid.go, bench_concurrent_runner.go, corpus.go, and the cache-eviction helpers (cold read numbers are meaningless against a warm page cache). Carry the harness and methodology; leave the prototype's reader code behind.

Reuses from #771

internal/fullhistory/bench/ — the cobra subcommand pattern, the CSV/percentile sink, and the fixture/source plumbing. Benchmark inputs (a populated hot DB / frozen cold chunks) can be produced by bench-ingest itself.

Out of scope

• CI harness and baselines (Run full-history benchmark commands in CI to catch perf regressions #775 — blocked on this + Full-history benchmark commands — ingest across all data types #771).
• Ingestion benchmarks (Full-history benchmark commands — ingest across all data types #771).
• Scenario suite (Implement the RPC v2 performance benchmark suite (data profiles + test scenarios) #776).

Dependencies

• Full-history benchmark commands — ingest across all data types #771's bench package (subcommand plumbing + CSV sink) — PR pending.
• Production cold/hot readers in storage/stores/ (landed).