[CI] Run the cuVS Bench Lucene E2E suite with provisioned artifacts

Problem

The cuVS Bench Lucene integration suite added in #2624 is intentionally selected with --run-lucene-e2e. Current Python CI runs ordinary pytest without that option and does not provision the custom PyLucene 10.2/JCC runtime, JDK 22, matching cuvs-java and thin cuvs-lucene JARs, or their native runtime. The live cases therefore remain manual evidence rather than continuous regression coverage.

Proposed change

• Decide whether the live suite should be blocking for relevant paths, scheduled, or both.
• Provision a reproducible PyLucene 10.2/JCC runtime and matching native dependencies.
• Pass exact cuvs-java and thin cuvs-lucene producer artifacts into the cuVS Bench job and record their revisions or hashes.
• Invoke the suite with --run-lucene-e2e.
• Fail clearly on missing prerequisites, unavailable GPU/cuVS paths, unexpected CPU fallback, or zero executed live cases.
• Report selected, passed, failed, and skipped case counts.

This requires coordination across cuVS Bench/Python, Java/Lucene artifact, and CI/packaging ownership. Until that handoff exists, the suite should remain documented as manual-only in current CI.

Start by locating the current Python CI and cuVS Bench job, then trace how --run-lucene-e2e is selected and how producer artifacts are passed. Review the required PyLucene/JCC runtime, JDK 22, cuvs-java, thin cuvs-lucene, and native dependencies. Done means reproducible provisioning, explicit prerequisite failures, live-case counts, and documented blocking or scheduled behavior.