perf: replace B-tree CONTAINS scan with fulltext index for search()

[aksOps]

Summary

Fixes RAN-66.

B-tree indexes on label_lower/fqn_lower cannot serve CONTAINS predicates in Neo4j — every call to search() caused a full graph scan regardless of the index. This replaces the CONTAINS queries with a fulltext index backed by Lucene.

• Fulltext index (search_index) created on (n.label_lower, n.fqn_lower) using the keyword analyzer — preserves whole-property tokens so FQNs with dots aren't split by the standard tokeniser
• GraphStore.bulkSave() — creates search_index alongside the existing B-tree indexes (used by the serve-path bootstrap)
• EnrichCommand — creates search_index in the secondary-index block and adds CALL db.awaitIndexes(300) so the index is fully ready before the first query
• search(String, int) and search(String) — use db.index.fulltext.queryNodes() with *text* wildcard wrapping for substring matching
• toLuceneQuery() helper — lowercases input and escapes Lucene special characters before wrapping

Test plan

• GraphStoreTest#shouldSearch — passes (mocked, verifies plumbing)
• GraphStoreExtendedTest#shouldSearchWithLimit — passes
• EnrichCommandTest — all 3 tests pass
• Full suite: mvn test — 1473 tests, 0 failures

🤖 Generated with Claude Code

[aksOps]

Code review

Found 2 issues:

1. EnrichCommand does not write label_lower/fqn_lower — search_index will index empty fields, breaking all search results

The fulltext index is created on [n.label_lower, n.fqn_lower], but the UNWIND bulk-load path in EnrichCommand.enrichFromCache() only writes label and fqn to the node property map — the lowercased variants are never populated. Every search via the standard index → enrich → serve pipeline will return zero results because the indexed fields are empty on all nodes. The GraphStore.bulkSave() path correctly writes label_lower/fqn_lower (lines 191-192), but EnrichCommand bypasses bulkSave() and its UNWIND batch does not include these properties.

https://github.com/RandomCodeSpace/code-iq/blob/4c3310a08c44c303361c52a432087d065b51b9ec/src/main/java/io/github/randomcodespace/iq/cli/EnrichCommand.java#L194-L200

Fix: add label_lower and fqn_lower to the node props map in EnrichCommand before the UNWIND batch, mirroring what GraphStore.nodeToProps() does at lines 191-192.

2. toLuceneQuery() does not escape * and ? — user input with wildcards is misinterpreted

The escape list covers 15 Lucene special characters but omits * and ? (the wildcard operators). A search for foo*bar or name? will be passed through to Lucene as multi-wildcard and single-character-wildcard queries respectively instead of matching the literal characters. Both should be escaped before the *...* wrapping is applied.

https://github.com/RandomCodeSpace/code-iq/blob/4c3310a08c44c303361c52a432087d065b51b9ec/src/main/java/io/github/randomcodespace/iq/graph/GraphStore.java#L272-L291

Fix: add .replace("*", "\\*").replace("?", "\\?") to the escape chain in toLuceneQuery().

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[aksOps]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

🤖 Generated with Claude Code

[aksOps]

Closing as chain/review PR — superseded by #16 which targets main directly. All changes from this branch are now in main.