Keep projected rows and readiness metadata consistent across JSON modes

[Widthdom]

Summary

Projection and JSON-shape selection change the meaning and placement of query results. Projected symbols/files calls can return results:[] while placing the first row under metadata.response_context, and symbols --json=array drops exact_index_available that is present in streaming JSON.

Audit IDs: Q-01 and Q-30. Reproduced with v1.40.3 at 404db09489c305f17d94f37026e00564d3ac527a. Closed #1527 is adjacent response/freshness contract work.

Expected behavior

Array, NDJSON, and envelope modes must serialize the same row DTO and readiness facts. Projection may remove only fields the caller did not request; it must not relocate rows or fabricate has_more:false.

PR scope

One PR should fix row projection/serialization across symbols and files. Unknown-field validation/schema discovery is tracked separately.

Implementation guide for Codex 5.6 sol xhigh

• Add a matrix covering symbols/files, projected/unprojected rows, array/NDJSON/envelope, empty/non-empty, and bounded pages.
• Build the projected row once, then pass it to format-specific writers; never reuse the first row as response-context metadata.
• Keep readiness/pagination metadata outside projected row fields and consistent across modes.
• Preserve backward-compatible field names, add contract tests, run net8/net9, and add a bilingual changelog fragment.

Acceptance criteria

• Every non-empty projected query returns all rows in the documented collection/stream.
• exact_index_available, counts, truncation, and cursor facts agree across JSON modes.
• No row appears under metadata.response_context.
• Empty results remain genuinely empty.

[Widthdom]

Work started on branch fix-issue4835.