fix(lower): topk over a bare selector preserves by-labels (#30)

[zzylol]

What & why

Fixes #30 — a follow-up to PR #18 (which removed the L3 Partition node and routed per-group ranking onto Sort.partition_by).

topk(k, <bare instant selector>) by (labels) — e.g. topk(3, http_requests_total) by (job) ("top-3 series per job") — mislowered:

topk(3, http_requests_total) by (job)
→ column resolution failed: column `job` not found in schema (have: ["sum"])

The non-heavy-hitter path in build's Outer::TopK else-branch defaulted a bare-selector argument (inner.func == None) to an implicit cross-series AggFunc::Sum. That reducing Sum collapsed every label (including the by partition keys), so Sort.partition_by = [job] no longer resolved at L3. It was also semantically wrong — PromQL topk ranks the raw instant-vector samples, it does not sum them.

How

A bare selector now ranks over the filtered_source directly (label-preserving), so Sort.partition_by ranks within each group:

topk(3, http_requests_total) by (job)
→ Limit{3} → Sort{ keys:[value desc], partition_by:[job] } → Scan(http_requests_total)

A range-vector-function argument (topk(k, rate(m[5m])), topk by (host) (k, avg_over_time(...))) still reduces per series first — also label-preserving — so those paths are unchanged, as are the heavy-hitter count_over_time and bottomk cases.

Tests

• topk_over_bare_selector_by_label_ranks_per_group — Sort.partition_by = [job] over a bare Scan, no implicit Sum.
• topk_over_bare_selector_ranks_raw_samples — no by → global ranking over raw samples.
• Full workspace suite green; cargo clippy --all-targets clean.

Closes #30. Follow-up to #18.

🤖 Generated with Claude Code