feat(sketch_algebra): implement Matcher for asap_plan::Implementation

[zzylol]

Summary

Step 2 of the sketch-identity unification plan (design doc — not committed to the repo, see conversation for the published artifact). Step 1 (audit whether ASAPController needs serde on SummaryKind/SummaryParams) concluded no — every real emit path in control_plane builds JSON/YAML by hand (build_backend_aggregation_json etc.), never via serde_json::to_value on a typed struct; the existing Serialize/Deserialize derives on sketch_algebra::SketchKind/SketchParams only back #[cfg(test)] round-trip tests. No ASAPController PR needed — folded into this repo's own work instead.

This step:

• Bumps the ASAPController pin to 283b6ab (merge of PR refactor(legacy_expr): step γ4 — TopK bridge (heavy-hitter vs Sort+Limit split) #141, "convert Implementation::is_satisfied_by into a Matcher trait") and adds asap-plan/asap-sketch as new control_plane dependencies (previously only asap-ir was pinned).
• Implements SummaryFamilyMatcher: Matcher — restores, verbatim, the family-compatibility logic that briefly lived as a concrete Implementation::is_satisfied_by method in asap-plan (PR refactor(legacy_expr): step γ1 — Aggregate bridge module + output_schema helper #140) before it became a trait with no default impl (PR refactor(legacy_expr): step γ4 — TopK bridge (heavy-hitter vs Sort+Limit split) #141). Pulled directly from ASAPController's git history (commit da8fa75) rather than reconstructed from memory, so all 10 ported tests are a faithful restoration.

Scope note: this deliberately does not implement the single-vs-multi-population re-aggregation rule (e.g. a keyed Sum accumulator serving an unkeyed Sum query) — Implementation carries no grouping information at all, so a two-Implementation Matcher can't correctly answer that question. That logic already exists as production code in asap_types::capability_matching::find_compatible_aggregation, which checks type compatibility and grouping-label subset-compatibility side by side; the two checks compose at the caller, not inside a single Matcher::is_satisfied_by.

This PR is purely additive — no existing enum (Capability, SketchKindHandle, sketch_algebra::SketchKind) is touched or deleted. That's later steps in the same plan.

Test plan

• cargo build --workspace — clean
• cargo test -p control_plane — 834 passed (824 + 10 new), 1 pre-existing unrelated failure (invalid_sketch_type_override_falls_back_to_default)

🤖 Generated with Claude Code