feat(summary_executor): read out named-key PointCount (count(metric{item=x}))

[zzylol]

Summary

Phase 4 of 5 toward supporting named-key SketchQuery::PointCount (count(cms_metric{item="x"})). Needs SketchQuery::PointCount.value from ASAPController#162 (merged) / #413 (merged) — both landed, so this is now based directly on main.

• Bumps data_plane's ASAPController rev pin to the merged 64df20d (matching feat(sketch_algebra): realize Frequency (AggIntent::Extension) as a real Cms sketch #413's control_plane pin) to pick up SketchQuery::PointCount.value: Option<String>.
• New SummaryState::estimate(key) in delta_apply.rs, mirroring topk_items's per-variant dispatch pattern but broader: all four Frequency variants (Cms, CountSketch, CmsWithHeap, CountSketchWithHeap) already carry a keyed estimate over their matrix — no heap needed for a point lookup, unlike TopK.
• sketch_query_value's PointCount arms now handle both shapes: {key: SampleValue, value: None} (bare total, unchanged behavior) and {key: Named/Qualified, value: Some(v)} (per-item point lookup via SummaryState::estimate), erroring on any other key/value combination rather than guessing.

Scope note: this closes the data_plane-side readout logic only. Threading the actual filter value into whatever constructs SketchQuery::PointCount in the first place (Phase 3 — control_plane's frequency() intent construction) turned out to need substantially more plumbing than expected: recognizing a bare count(metric{item="x"}) PromQL shape and resolving item against a metric's configured item-label requires a new metric→item_label lookup mechanism threaded through parse_query_expr_canonical (18 call sites) and QueryWorkload (31 struct-literal construction sites, no Default impl) — a genuinely larger, separate task, deferred for now. This PR's new tests exercise the readout logic directly with hand-built SketchQuery::PointCount{value: Some(_)} fixtures instead of an end-to-end PromQL string, which is exactly the shape Phase 3 will eventually produce.

Test plan

• cargo test -p data_plane --lib summary_executor — 17 tests (15 existing + 2 new), all passing:
  • single_cms_sid_named_key_point_estimate — a single sid, point lookup for a specific key, asserts the exact inserted weight.
  • two_cms_sids_same_group_named_key_estimate_merges_cross_sid — two sids contributing to the same key, proves the merged answer is the sum (the cross-sid merge proof, mirroring two_cms_sids_same_group_totals_actually_merge).
• cargo test -p data_plane --lib — full lib suite: 913 passed, 2 ignored.
• cargo test -p data_plane --test e2e_controller_plans_and_backend_serves — 2 pre-existing failures (controller_plan_to_query_full_roundtrip_{cms,count_sketch}_with_heap_topk), confirmed identical on the unmodified branch via git stash before/after comparison — not a regression from this change.
• cargo build --workspace — clean.
• cargo clippy -p data_plane --lib --tests — 107 warnings, same count as this branch's baseline; verified none fall in the touched files.
• rustfmt scoped to the two touched files.

🤖 Generated with Claude Code