feat(search): migrate unified search from Elasticsearch to PostgreSQL FTS pipeline

[hyunbin1]

Motivation

• Replace the Elasticsearch-based unified search pipeline with a self-contained PostgreSQL Full-Text Search implementation while keeping existing API signatures to minimize frontend changes.
• Preserve existing domain preprocessing and ranking intent but map ES boost/signals to tsvector weights and SQL ranking formulas suitable for Korean text and operational constraints.
• Keep Redis as a caching/fallback layer and provide a migration path (DDL, triggers, and rebuild) to remove ES from infra later.

Description

• Added a new JDBC-backed repository PostgresUnifiedSearchRepository that implements unified search, autocomplete and top-keywords using PostgreSQL FTS and ranking SQL; file: src/main/java/nova/mjs/domain/thingo/ElasticSearch/Repository/PostgresUnifiedSearchRepository.java.
• Switched service wiring so UnifiedSearchService now delegates to the Postgres repository while preserving the existing search(keyword, type, category, order, pageable) API; file: src/main/java/nova/mjs/domain/thingo/ElasticSearch/Service/UnifiedSearchService.java.
• Reworked SearchIndexSyncService to build a full thingo_search_document table in Postgres from domain repositories and refresh FTS vectors instead of creating ES indices; file: src/main/java/nova/mjs/domain/thingo/ElasticSearch/Service/SearchIndexSyncService.java.
• Converted SuggestService to call the Postgres autocomplete SQL ranking (personal/global/time-decay) and updated RealtimeKeywordService to persist search logs to Postgres while retaining Redis ZSET behavior for hot cache/fallback; files: SuggestService.java, RealtimeKeywordService.java.
• Added Postgres schema, indexes, trigger and materialized view SQL for FTS, autocomplete and realtime keywords at src/main/resources/db/search/postgres_search_schema.sql and documented the migration, architecture, ranking and verification plan in docs/postgresql-search-migration-plan.md.

Testing

• Attempted a build check with ./gradlew compileJava to validate compilation, but the build could not run in this environment due to a Gradle/JDK incompatibility reporting Unsupported class file major version 69 (build did not complete).
• No automated unit/integration tests were executed in this environment after the changes; manual validation and CI build are required to confirm compilation and runtime behavior against the target JDK and Postgres 15+ environment.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6bf2d022e9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Guard search-log writes so detail search still succeeds

recordSearch now performs a PostgreSQL insert on the request path without handling failures. If search_query_log is not yet migrated or the DB write fails transiently, this exception bubbles out of /api/v1/search/detail after search results are computed, turning successful searches into 500s. This logging path should be best-effort (catch/log and continue) so analytics outages do not break user-facing search.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Make table rebuild atomic to avoid empty search index

rebuildSearchDocuments truncates thingo_search_document before repopulating it, but this destructive step is not wrapped in a transaction with the subsequent batch insert. Any mid-rebuild failure (constraint error, timeout, connection drop) can leave the table empty or partially loaded, causing a broad search outage until a manual rerun. The truncate+insert path should run atomically (or use a staged swap).

Useful? React with 👍 / 👎.