search --type bm25 fails on tables with a column named score (duplicate relevance column) #179

Summary

hotdata search --type bm25 fails when the target table has a user column named score, because the BM25 relevance column emitted by bm25_search() is also named score, producing a duplicate-column collision before projection. Affected tables cannot be BM25-searched at all from the CLI.

Error

Schema contains duplicate qualified field name "bm25_search()".score

Repro

# 1. Managed DB with a table that has a column literally named `score`
hotdata databases create --name "bm25 score repro" --catalog bm25repro --table widgets
hotdata databases set <db-id>
# widgets parquet has columns: id, name (text), score (double)
hotdata databases load --catalog bm25repro --table widgets --file widgets.parquet

# 2. BM25 index on the text column
hotdata indexes create --catalog bm25repro --table widgets --column name --type bm25

# 3. Search -> collision
hotdata search "alpha" --type bm25 --table bm25repro.public.widgets --column name
# error: Schema contains duplicate qualified field name "bm25_search()".score

Notes

• --select id,name does not avoid it — the collision happens in the generated query's schema before the projection is applied.
• Confirmed BM25 search works fine on a table without a score column (e.g. the listings demo table), where the relevance column is surfaced as score. That confirms the user's score column is the culprit.

Expected

A table that happens to have a score column should still be BM25-searchable. Options:

• Alias the relevance column to a non-colliding / namespaced name (e.g. _bm25_score or bm25_score), or
• Detect the collision and auto-suffix the generated relevance column.

Environment

• hotdata-cli 0.7.0 (built from main @ b250d0a)
• API: https://api.hotdata.dev/v1
• Found during full E2E verification of the CLI against production.