Package and validate Lucene backend

[nvzm123]

Note

Draft 4 of 5 in the Lucene backend split. It activates and packages the
backend implemented by the preceding stack layers.

Important

The independent cuvs-lucene query-matrix cleanup PR
#16 is a merge prerequisite for
activating CAGRA search through this backend.

Summary

• register the lucene backend and configuration-loader entry points;
• verify Lucene algorithm resources and plugin metadata in built packages;
• add opt-in live CPU/GPU integration coverage with path-specific fallback checks; and
• keep ordinary test runs independent of the external PyLucene/JVM/GPU runtime.

Validation

• focused Lucene tests: 158 passed, 13 skipped;
• opt-in live PyLucene CPU/GPU suite on an NVIDIA A10G: 13 passed;
• selected cuVS Bench Python suite: 329 passed, 37 skipped, 4 deselected;
• isolated setuptools-fallback wheel build, install, entry-point discovery,
  and backend construction: passed;
• Python compilation and git diff --check: passed.

The 13 ordinary-run skips require --run-lucene-e2e; the remaining 24 selected-
suite skips cover unavailable optional dependencies or services. The explicit
live run used the exact stack JAR with a preserved PyLucene 10.2 runtime and
matching cuvs-java/native 26.12 artifacts; PyLucene was not rebuilt from source.
Of the four deselections, two pre-existing C++ dry-run cases write beneath
read-only /data; the other two CLI fixture cases require benchmark outputs
unavailable in this container and fail identically on main.