Fix PyLucene sidecar packaging and add smoke test

[nvzm123]

This builds on and can supersede #147.

PR #147 adds PyLucene support, but GPU E2E testing exposed several packaging/runtime issues:

• the PyLucene sidecar merged Lucene-owned SPI descriptors, causing PyLucene to try loading incompatible Lucene providers
• the sidecar flattened cuvs-java, which breaks its multi-release JDK 22 provider classes
• PyLucene needs the base cuvs-java jar on the JVM classpath separately from the cuvs-lucene sidecar

This PR updates the sidecar to include only cuvs-lucene classes and service descriptors, removes Lucene-owned service providers from the cuvs-lucene SPI descriptor, documents the required classpath layout, and adds a PyLucene smoke-test wrapper.

Validation performed on a g5.2xlarge EC2 instance with NVIDIA A10G:

• built cuvs-lucene against cuVS 26.08
• built/installed PyLucene 10.0.0
• ran CuVS2510GPUSearchCodec through PyLucene
• indexed/searched 2,000 rows x 32 dimensions
• verified cuVS index files were produced (.vcag, .vemc)

Passing smoke output:

PASS: CuVS2510GPUSearchCodec indexed/searched 2000 rows x 32 dims via PyLucene

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.