Stage 3: evidence-driven cost (storage I/O, hand-off bytes) #615

Problem

Stage 3 (asap_plan_selection::plan_stages) ranks physical plans without storage-I/O or hand-off-byte evidence. The only tests showing that such evidence changes which plan is selected lived in the dag_export devtool bin, which ran them through the legacy PhysicalPlanCostModel (PlannerPhysicalPlanProvider + PhysicalEvidenceSnapshot). That bin is being deleted (Q50, approved 2026-10-04), and Stage 3 never used that cost path.

Coverage lost with crates/devtools/src/bin/dag_export.rs:

• storage_requests_export_and_change_plan_selection — a StorageIoProfile (disk/object requests per node) flips plan selection between a raw plan and a summary candidate.
• handoff_bytes_export_and_change_plan_selection — a PhysicalHandoffProfile (materialization / transfer bytes) flips plan selection.
• cache_json_affects_ranking_and_exports_declared_evidence — cache hit-ratio evidence changes ranking.
• global_selection_uses_the_cheapest_complete_physical_candidate, missing_physical_evidence_keeps_the_export_raw_only, cost_annotations_fail_closed_without_physical_evidence, incomplete_or_unused_json_evidence_fails_closed, duplicate_target_candidate_and_query_evidence_each_fail_closed — evidence-ranked physical cost and its fail-closed behaviour.

The unit tests for the evidence types themselves (asap-plan-selection --test storage_io, --test physical_handoff_cost) remain, but nothing end-to-end checks that evidence affects selection.

Wanted

Stage 3 cost takes storage-I/O and hand-off-byte evidence into account (or explicitly decides not to), with tests that show the evidence changing the selected plan through plan_stages.