Show deployment inputs in the stage document and viewers

[zzylol]

Stack: Wave 3 chain: #605 → #607 → #608 → #609 → #610

Rebased on main d4869a7 (DF 54).

Why

#609 adds the deployment inputs to Stage 3, but the asap-stage-pipeline/v1 document does not show which inputs a run used. A reader of a fixture cannot tell why a candidate was rejected for a capability, or which memory weight priced it (C4).

What

stage_pipeline writes a new, optional top-level deployment section:

"deployment": {
  "capabilities": {
    "source": "asap_executor::capabilities",
    "summaries": [ { "summary": "Kll", "readouts": ["Quantile"] },
                   { "summary": "HydraCms", "readouts": ["TotalCount", "ItemCount"] },
                   { "summary": "exact Sum", "readouts": [] }, ... ],   // null = unrestricted
    "ingestion_time": true,
    "query_time_retention": false,
    "memory_budget_bytes": null,
    "raw_data_retained": true,
    "raw_bytes_per_sample": 16
  },
  "cost_model": {
    "name": "analytical-cost-v2",
    "unit": "cost_per_second",
    "calibration": { "version": "illustrative-v2", "cost_per_cpu_op": 1e-6,
                     "cost_per_scan_byte": 1e-7, "cost_per_retained_byte_second": 1.25e-7,
                     "horizon_s": 3600.0, "latency_ms_per_cost_unit": 1.0 }
  },
  "accuracy_model": { "name": "DefaultAccuracyModel", "evidence": "none" }
}

Supporting changes: SummarySupport::name() ("HydraCms", "exact Sum"), and asap_plan_selection::COST_MODEL, which candidate cost source strings now also use.

How (Before / After)

• Before: the document has workload and stage0–stage3; the inputs are implicit.
• After: it also has deployment. The Example 1 fixture is regenerated, and only this section is added (+100 lines). The 4a document also generates with the section; its fixture lives on feat(dag-viewer): redesign the viewer around the stage pipeline #574.
• Viewer: tools/dag-viewer/stages.js / validateStagePipeline is not on this branch, so nothing here validates top-level keys; python3 -m unittest test_render passes (29). The section is optional, so the feat(dag-viewer): redesign the viewer around the stage pipeline #574 validator should accept it when present and not require it.
• What the viewer's "Deployment inputs" panel should show (to port onto feat(dag-viewer): redesign the viewer around the stage pipeline #574):
  1. Capabilities: a summary × readouts table (or "unrestricted" when summaries is null), plus flags for ingestion time, query-time retention and raw data retained, the memory budget, and bytes per raw sample.
  2. Cost model: name, unit, calibration version, and the five coefficients, with the memory weight (cost_per_retained_byte_second) highlighted because it also prices raw retention.
  3. Accuracy model: name and evidence.
  4. Optionally, link Stage 3 rejections whose reason starts with deployment (or contains memory budget) to this panel.

Gate: cargo fmt --check, clippy -D warnings, cargo test --workspace: 1,662 passed, 23 ignored (#609: 1,661 / 23). There is a new devtool test for the section, and the devtool test's temp-file names are made unique, because two tests now generate the same document in parallel.

Stacked on #609. Links: #509, #525, #572, #574, #594, #604.

🤖 Generated with Claude Code