feat(search): match Elasticsearch token semantics exactly (PPT-2644)

[camreeves]

Part of PPT-2644. Follow-up to #445, restoring exact Elasticsearch token semantics for multi-word q to avoid behavioral regressions against the pre-migration platform.

What changes

• Token joining: AND → OR. feat: replace Elasticsearch with PostgreSQL full-text search (PPT-2644, 2/2) #445 shipped AND-of-prefixes (every term required). This returns to what ES's simple_query_string did: a record matches when any term matches.
• Prefix scope: last token only. ES appended * to the query's final token; earlier terms were whole-word matches. Now identical. (sydney roo = whole-word sydney OR prefix roo — the term being typed still autocompletes.)
• Email precision preserved. Under pure OR, q=<address> would OR-split and match every user on the domain. Old clients had precise email lookup via the (now accepted-and-ignored) fields=email phrase path, so email-shaped input is kept as a single whole-address lexeme — the vectors index addresses both whole and tokenized for exactly this.
• New dedicated Utils::TextSearch spec (nine cases: the never-500 vectors, ES-syntax degradation, email handling, token/length caps, unicode).

Notes for reviewers

• Multi-word searches will return more results than feat: replace Elasticsearch with PostgreSQL full-text search (PPT-2644, 2/2) #445's AND behavior (and identical-or-slightly-more than old ES, since nothing else narrowed) — X-Total-Count grows accordingly. This is the intended ES parity.
• The sanitizer's never-500 guarantee is unchanged: ES operator syntax, field prefixes, quotes and junk all degrade to plain terms.
• Full containerised suite: 885 examples, 0 failures, 0 errors (1 pre-existing pending).

[greptile-apps]

Greptile Summary

This PR restores Elasticsearch-compatible token semantics for PostgreSQL text search while replacing placeholder-based email handling with collision-free tokenization.

• OR-joins search tokens and applies prefix matching only to the final token.
• Preserves email addresses as whole quoted lexemes.
• Adds coverage for sanitization, Unicode, limits, email handling, and the previously reported placeholder collision.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
src/placeos-rest-api/utilities/text-search.cr	Restores OR-based Elasticsearch token behavior and replaces collision-prone email placeholders with ordered single-pass tokenization.
spec/text_search_spec.cr	Adds focused coverage for token joining, final-token prefixes, syntax degradation, emails, collision handling, Unicode, and resource caps.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A[Free-form query] --> B[Cap input length]
    B --> C[Single-pass tokenizer]
    C --> D{Token type}
    D -->|Email| E[Quoted whole-address lexeme]
    D -->|Field prefix| F[Discard]
    D -->|Operator| G[Discard]
    D -->|Plain word| H[Word lexeme]
    E --> I[Cap token count]
    H --> I
    I --> J[OR-join tokens]
    J --> K[Prefix final token only]

Loading

_{Reviews (2): Last reviewed commit: "refactor(search): tokenize in a single o..." | Re-trigger Greptile}