fix(pgwire): keep the planner's output keys on the extended protocol

[EnRaiha]

Why

Over Parse/Bind/Execute (extended protocol), two output columns sharing a name render the same cell: SELECT w.id, b.id … returns (w1, w1) and SELECT nextval('s'), nextval('s') returns (1, 1).

The Describe phase built its OutputSchema with lookup_key = display_name = field name (prepared/execute.rs:136-149), and the execute path replaced the planner's columns with it (routing/execute.rs:153-160). Rows are written under the unique keys cell_keys derives (id, id_1), so the second column's cell was never read. Simple query has been correct since the derived-key work.

Fixes #337.

What

• Merge instead of replace: lookup_key comes from the planner's build_output_schema — the single derivation that runs cell_keys; the announced schema supplies display_name, ty, and is_star; cp_computed stays the planner's.
• An announced arity that disagrees with the plan keeps the announced columns but derives their keys with cell_keys, so the reader still agrees with the row writer.
• effective_output_schema is private, unit-tested in place; no signature or public-API change.

Validation

• Wire (extended protocol) on the harness: SELECT 1 AS id, 2 AS id → (1, 2); SELECT w.id, b.id FROM w JOIN b … → (w1, b1). Both cases fail on the current main (red: left: 1, right: 2; the join case reports error retrieving column 0) and pass with the fix (green).
• cargo check -p nodedb --all-targets — clean.
• cargo test -p nodedb --lib routing::execute — pass (merge keeps the planner keys; arity mismatch derives announced keys).
• rustfmt clean; the repository preflight passes (test attributes +2).

Notes

• Presentation-only: no authorization, redaction, or audit change; display names, types, and result formats are unchanged.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[farhan-syah]

Closing: right behaviour, wrong place.

The producer is prepared/execute.rs — the Describe phase builds lookup_key: f.name() instead of deriving keys with cell_keys(names). Fix that one site and the existing replace-at-execute stays correct. The 40-line downstream merge is not needed; its arity-mismatch branch already is the producer fix. Resubmit as the producer fix plus the wire test.