Tracking: correct AtomicCosts table for ASAPQuery#524 (parent: ASAPQuery#523) #32

Tracking issue for the sketch-bench-side work needed before ASAPQuery#524 (part of ASAPQuery#523's greedy-optimizer TODOs) can load a correct per-(sketch_type, params) atomic-costs table.

Dependency chain: this issue → ASAPQuery#524 → ASAPQuery#523.

• Emit a derived AtomicCosts table for the ASAPQuery optimizer #30 — merge/subtract timing (subtract partly blocked on asap_sketchlib#69), cardinality sweep, export/derivation script, grid alignment with ASAPQuery's candidate_gen.rs
• Memory measurement: add per-sketch heap tracking; tighten existing metric semantics #14 — memory-metric correctness (HLL register-width bug, jemalloc field/semantics mismatch, ExactFrequency memory undercount) — blocks the mem column
• Add CPU-time instrumentation for the query phase (query_cpu) #31 — query-phase CPU-time instrumentation — blocks the query_cpu column
• merge_passes: silently drops conflicting records instead of surfacing duplicates #28 — merge_passes silently drops conflicting records instead of surfacing duplicates — relevant if the export script needs to merge multiple sweep passes into one table