fix(platform): stop a search returning one passage twice

[Israeltheminer]

A search no longer spends two of its result slots on the same passage.

Closes #3061. Supersedes #3105, whose two files #3125 deleted.

Why

The same text can sit in the corpus twice — one file uploaded as two documents, or a paragraph two documents share. Both copies match, and a bounded result set then says one thing twice while a different answer falls off the end. A caller asking for eight passages can receive seven distinct ones.

Observed as duplication on a deployment: two pairs of corpus rows with matching content hashes, chunk counts and character counts under different refs.

What changed

dropRepeatedPassages in backend/core/knowledge/search.ts, keyed on the text a caller actually reads.

The corpus is part of the key, so the same wording found in a document and on a crawled page stays two answers — they are different claims about where it came from. Whitespace is normalized, so two copies differing only in how their source wrapped lines count as one. Fusion has already sorted by score, so the first occurrence is the best one and no comparison is needed.

The order is load-bearing. It runs after the retrievability filter. Deduping first could keep a copy the caller cannot read and drop the readable one; the gate would then remove what was kept, losing the passage entirely rather than showing it once.

Tests

Five cases: two identical copies collapse to the higher-scoring ref, two distinct chunks of one document both survive, two copies differing only in line wrapping collapse, the readable copy is kept when the other is filtered out, and the same wording in two corpora stays two hits.

Mutation	Went red
dedupe not called	2 cases
whitespace normalization dropped	the wrapping case
corpus dropped from the key	the cross-corpus case
dedupe moved before the retrievability gate	the readable-copy case

The last one is the reason that case exists — the ordering bug is invisible until a copy is unreadable.

Scope

Nothing purges the duplicate corpus rows themselves. This stops them being served twice; the rows remain, and a replacement upload still leaves an orphaned row behind. That is tracked separately in #3142.

Gate: typecheck, oxlint --type-aware, oxfmt --check, lint:sast green; branched from origin/main at e616480a8.