Add Lucene benchmark backend core

[nvzm123]

Note

Draft 3 of 5 in the Lucene backend split. It depends on the generic plugin
contract and PyLucene runtime layers below it.

Summary

• add the directly importable Lucene cuVS Bench backend and configuration loader;
• support Lucene CPU HNSW, cuVS-built HNSW, and cuVS CAGRA search;
• record explicit build/search timing and provenance boundaries;
• validate algorithm parameters, persisted codecs, vector fields, and CAGRA indexes;
• keep fallible validation and diagnostics before rollback-safe index publication;
• cover correctness, scoring, lifecycle, rollback, publication, and result metadata.

This layer deliberately does not register public package entry points yet, so
its behavior can be reviewed and unit-tested before package activation.

Validation

• backend contract, build/search, lifecycle, result, and inherited runtime tests:
  152 passed;
• Python compilation and git diff --check: passed.