[Feature] Add happy-path execution tests for vectorSearch()

[mengweieric]

Summary

• Adds VectorSearchExecutionIT with end-to-end execution coverage for top-k, POST-filter, EFFICIENT-filter, and both radial modes (max_distance / min_score). VectorSearchIT already covers rejection paths; this adds conditional happy-path coverage that runs locally against a k-NN-enabled cluster.
• The k-NN plugin is not provisioned by the default integ-test cluster, so each test guards init() with Assume.assumeTrue(isKnnPluginInstalled()). Tests skip cleanly when k-NN is absent and run when it is. Ongoing CI regression protection depends on a separate follow-up to provision k-NN in the integ-test cluster.
• Test fixture is a 6-doc 2D knn_vector index with two well-separated clusters, pinned to Lucene HNSW + L2 so efficient filtering is deterministic and L2 scoring is well-defined.

What the tests assert

• Top-k — all 3 returned docs come from the near cluster; scores are non-increasing.
• POST filter — vector from far cluster + WHERE state='TX' with k=10 returns exactly the 3 near-cluster docs {1,2,3} (proves the filter is applied, not just vector similarity).
• EFFICIENT filter — vector near TX + WHERE state='CA' with k=3,filter_type=efficient returns exactly the 3 CA docs {4,5,6}. This is discriminative: a POST-filter implementation at k=3 would return 0 rows (all 3 nearest candidates are TX and get filtered out), so only an efficient-filter path that counts matching neighbors toward k produces this result.
• Radial max_distance — returns exactly {1,2,3} (cluster A within L2 distance 1.0).
• Radial min_score — returned docs satisfy the score floor and match exactly {1,2,3}.

Follow-up

Provisioning k-NN in integ-test CI is a separate change tracked outside this PR; once k-NN is on the test cluster the skip becomes a no-op and these tests light up automatically.

Test plan

• ./gradlew :integ-test:integTest --tests "org.opensearch.sql.sql.VectorSearchExecutionIT" — 5 skipped on default cluster (no k-NN), 0 failures
• ./gradlew spotlessCheck
• Run locally with k-NN installed to confirm all 5 tests pass