feat(intent_algebra): thread the actual item filter value into Frequency (Phase 3a)

[zzylol]

Summary

Phase 3 (of 5) of the named-key SketchQuery::PointCount gap — workload-driven half only. Based directly on main (#162/#413/#414 already merged).

Split from the original Phase 3 scope after investigation: the PromQL-string-driven half (recognizing a bare count(metric{item="x"}) and resolving item against a metric's configured item-label at parse time) needs a new metric→item_label lookup threaded through parse_query_expr_canonical (18 call sites) and QueryWorkload (31 struct-literal sites, no Default) — a genuinely larger, separate task, still deferred. This PR does the achievable, low-risk half.

• frequency(accuracy, item: Option<(String, String)>) now embeds item_label/item_value into the Extension payload when a specific item is being filtered on. ControlPlaneCostModel::readout_extension (from feat(sketch_algebra): realize Frequency (AggIntent::Extension) as a real Cms sketch #413) already reads these exact keys back out — this closes the other half of that wiring, not a new payload shape.
• default_frequency() stays 0-arg (passes None), so the two PromQL-driven lower.rs call sites (agg_func_to_intents) are untouched — only the 4 other direct frequency() callers (3 tests + 1 real) needed updating for the new parameter.
• bind_workload_typed gains a sibling, bind_workload_typed_with_item_filter, rather than adding an item_label field to QueryWorkload itself (which would break all 30+ struct-literal construction sites, no Default impl to fall back on).
• Wired at the one real call site that already has everything needed in scope with zero new lookups: emit::collect_metric_to_family's loop already walks WorkloadEntry (has item_label) and QueryWorkload (has label_filters) together.

Test plan

• cargo test -p control_plane --lib — 766 passed, 1 failed (the same single pre-existing, unrelated failure present on unmodified main).
• New test bind_workload_typed_with_item_filter_threads_the_actual_value — proves both directions: no filter still reads out as the bare bucket total (PointCount{key: SampleValue, value: None}, unchanged), and a real (label, value) filter reads out as PointCount{key: Named(label), value: Some(value)}.
• cargo build --workspace — clean.
• cargo clippy -p control_plane --lib --tests — verified every warning falls outside this diff's touched line ranges (all pre-existing).
• rustfmt scoped to touched files only.

Remaining deferred: the PromQL-string bare-count() recognition half, and the overarching goal (wiring SummaryExecutor into the live serving path, retiring sketch_reducer.rs).

🤖 Generated with Claude Code