Read-endpoint SLAs under ingestion load #870

TL;DR

Read endpoints must hold their latency SLAs while the node is ingesting at target load — not just against a quiet store. Today the ingest benchmarks (#771) and query benchmarks (#856) run separately; this issue combines them into one scenario.

Scope

• A combined scenario: paced hot ingestion running at a phase's block cadence (--close-interval = 2 s / 1 s / 600 ms for phases 1/2/3 as defined in Synthetic ledger datasets for Phases 1–3 ("golden packs") #867) with the query benchmarks firing concurrently against the same node.
• Input datasets per phase (Synthetic ledger datasets for Phases 1–3 ("golden packs") #867), so both the write pressure and the data shape match the phase being claimed.
• Sweep query concurrency; measure read p50/p99 per endpoint (getTransaction, getEvents, getLedgers, getTransactions equivalents in the full-history store).
• Report interference explicitly, in both directions: reads-under-load vs the quiet-store baseline from Full-history query benchmark commands — reads across all data types  #856, and ingestion pace-lag with and without read load.

Done when

For each phase in scope, a pass/fail statement per read endpoint under ingestion load, reproducible as a campaign (ideally runnable through #868).