DAG viewer: draw the memo after each step (groups with their candidates) #1005

Problem, by example

Example workload: one query, evaluated every minute, with max_error 0.01 and failure_probability 0.001:

avg by (job) (quantile_over_time(0.5, data[5m]) / quantile_over_time(0.5, data[5m]) + quantile_over_time(0.9, data[5m]))

The memo the production planner builds for it, read from the stage-pipeline document tools/dag-viewer/examples/memo-ratio-of-quantiles.json (main at d728d05c, made by stage_pipeline from tools/dag-viewer/memo-examples/promql/ratio-of-quantiles-5m.json):

Corrected 2026-10-28: production numbers from the stage-pipeline document; earlier numbers came from a test helper without the series-identity column.

The two quantile_over_time(0.5, data[5m]) are one group, G2: share_common_sub_dags (crates/types/src/ir/cse.rs) shares a non-ASAP node only when its schema has a provable unique key, and the PromQL frontend's series-identity column ($promql_series_identity) gives the per-series quantile's output one (series identity, timestamp). A non-ASAP node without a unique key, such as a without(..) grouping, would still be one group per reader.

The DAG viewer can't show any of this. It draws three lanes for one plan: Stage 0, the selected Stage 1 candidate and its Stage 2 DAG. A reviewer can't see:

• that Stage 0 made 7 groups (the two quantile 0.5 share G2);
• which 7 candidates Pass 1 gave each quantile group;
• which 336 Panes and 345 SharedBuild candidates Pass 2 and the concrete-state step added.

So a missing rule, a wrong rule or a missed sharing can't be found from the viewer.

Proposal

A Memo view in tools/dag-viewer, reading the document's memo section:

• Stage tabs: Stage 0, Pass 1, Pass 2, concrete states.
• Groups: each group is a cytoscape compound node with its candidates inside. An edge goes from a candidate to each child group it reads.
• Candidates: colored by origin. A badge marks the candidates this step added. The candidates in selected_choice are highlighted.
• Details: clicking a candidate shows its own nodes in the details panel.
• Large memos: groups with more than N candidates collapse to one count per origin (336 Panes), and expand on click.
• Highlight query: the existing menu also dims, in the Memo view, the groups that query doesn't reach.

Scope and acceptance

• stages.test.mjs covers the memo validation and the per-step diff.
• test_viewer.py checks, on the example above, 7 groups on the Stage 0 tab and 19 candidates on the Pass 1 tab, and that the Pass 2 tab shows G2's Panes and SharedBuild counts.
• The example above is added to examples.json.

Depends on #1003.