Fix full-text-search recipe: correct /v1/search response shape

[claudespice]

Summary

Ran the full-text-search recipe end to end on current stable. The configuration and the
text_search() SQL path are correct, but the documented /v1/search response does not match
what the endpoint returns.

What the recipe said — a data object holding the requested path, and matches as a string:

{
  "matches": { "content": "... Follow these steps to get started with ..." },
  "data": { "path": "postgres/rds/README.md" },
  "primary_key": { "path": "postgres/rds/README.md" },
  "_score": 1.41,
  "dataset": "cookbook_files"
}

What the endpoint returns — no data key at all for this request, and matches maps each
indexed column to an array of matching text:

{
  "matches": { "content": ["# AWS RDS for PostgreSQL\n\nWorks with `v1.0+`\n\nFollow these steps to ge..."] },
  "primary_key": { "path": "postgres/rds/README.md" },
  "_score": 1.12,
  "dataset": "cookbook_files"
}

The reason is worth documenting rather than just patching the JSON: data carries only those
additional_columns that are not part of the primary key. Because path is this dataset's
full_text_search.row_id, requesting it returns the value under primary_key and emits no
data object. Requesting non-key columns does produce one — confirmed with
"additional_columns": ["size", "name"], which returns
"data": { "name": "README.md", "size": 2437 }. Added a short field-by-field list covering this.

Also corrected in the same pass:

• The first SQL example claimed LIMIT 5 but showed only 3 result rows, and its top hit was
  vectors/README.md — a path that does not exist in the repo (the recipe is vectors/s3).
  Replaced with the real five-row result and relabelled it "paths and scores vary", since the
  indexed source is the live cookbook repo.
• show tables; and the search result table now include the data-type row that spice sql prints.

Verified against

spiceai/spiceai at v2.1.4 (current stable)

Evidence

Ran with GITHUB_TOKEN set, per the recipe's prerequisites:

2026-08-06T12:18:14.265810Z  INFO runtime::init::dataset: Dataset cookbook_files registered (github:github.com/spiceai/cookbook/files/trunk), acceleration (arrow), results cache enabled.
2026-08-06T12:18:18.614284Z  INFO runtime::accelerated_table::refresh_task: Loaded 124 rows (1.33 MiB) for dataset cookbook_files in 4s 347ms.

$ curl -sS -X POST http://localhost:8090/v1/search -H 'Content-Type: application/json' \
    -d '{"datasets":["cookbook_files"],"text":"getting started","additional_columns":["path"],"limit":5}'
# result keys: ['matches', 'primary_key', '_score', 'dataset']      <- no "data"

$ # same call with "additional_columns": ["size","name"]
# result keys: ['matches', 'data', 'primary_key', '_score', 'dataset']
# data: {'name': 'README.md', 'size': 2437}

The text_search() SQL examples all ran successfully and _score is correct throughout.

[grokspice]

@copilot review