feat(asap-tools): support sub-second exact-MV E2E query intervals #876

Goal

Allow the sketch-bench E2E ClickHouse exact materialized-view (baseline_mv) arm to execute queries with a 100 ms repetition interval, so the five-system accelerated random-workload experiment can include the 100 ms spatial query class.

Current limitation

The merged #845 MV renderer uses integer interval_secs values and T-second buckets. It models an instant vector as a 1 s latest-sample window and admits an MV only when window_secs % interval_secs == 0. A workload with a 100 ms query interval therefore has no baseline_mv arm.

This is an experiment-runner/rendering limitation, not a ClickHouse limitation: the raw table already uses DateTime64(3).

Scope

• Carry query repetition intervals in milliseconds through the E2E table/config/client path while preserving current second-based inputs.
• Create and query baseline_mv buckets at millisecond resolution using DateTime64(3)-safe flooring and boundary expressions.
• Make instant-vector selection exact for the configured 100 ms scrape/lookback semantics.
• Preserve exact range semantics, including the closed PromQL left boundary and the existing edge-row handling.
• Update MV eligibility to use millisecond units.
• Add raw-vs-MV equivalence tests at 100 ms intervals, including instant selectors and at least one range query.
• Verify the generated ClickHouse client config can schedule a 100 ms query group and align explicit evaluation times.

Acceptance criteria

• A workload containing a 100 ms-query-cadence spatial query produces both baseline and baseline_mv ClickHouse configs.
• The MV and raw ClickHouse answers agree at fixed millisecond-aligned evaluation timestamps.
• Existing whole-second workloads and their generated SQL/configs remain supported.
• The five-system E2E workload preflight accepts the 100 ms spatial class.

Context

This unblocks the accelerated random-workload E2E plan: ten workloads, each with ten atomic queries, against ASAPQuery, Prometheus, VictoriaMetrics, raw ClickHouse, and ClickHouse+MV.