OPENNLP-1886: UniNE light and minimal stemmer tiers with vocabulary parity fixtures

[krickert]

Rebased on main; #1163 (StemmerFactory) is merged, so the diff is this change alone.

Adds sixteen stemmers in a new opennlp.tools.stemmer.light package: light and minimal tiers for German, French, Spanish, Norwegian (Bokmaal and Nynorsk varieties), and Swedish, plus light stemmers for Finnish, Hungarian, Italian, Portuguese, and Russian, and the English minimal stemmer. They fill the tier between no stemming and the Snowball algorithms: plural and inflectional endings are removed with far less conflation. Adapted from Apache Lucene's analysis-common module; the UniNE algorithms keep Jacques Savoy's BSD notice in each source file and in the distribution LICENSE.

Each stemmer is stateless and thread-safe, implements Stemmer, and is its own StemmerFactory. Input is expected lowercase, matching the algorithms' original contract; null input is rejected with IllegalArgumentException.

Parity is asserted against the original implementations: bundled fixtures sample the algorithms' vocabulary test data (up to 2000 word/stem pairs per stemmer, complete lists for Norwegian) regenerated by running the originals, so any behavioral drift fails the test. The Galician and Portuguese minimal stemmers are excluded because they build on the RSLP rule engine, which is its own effort.

Review round of 2026-09-04

Javadoc on every method and constructor, @throws on the validating paths, the StemmerFactory instance contract honored and pinned, the current Spanish plural rules with their sources cited, supplementary-plane input covered, and the contract and parity tests parameterized so a failure names the word.

Verification

opennlp-api and opennlp-runtime build with checkstyle and forbidden APIs; the stemmer test classes pass (346 tests, of which 181 are the light and minimal contract, parity, and usage tests), and the manual section's examples are mirrored by LightStemmerUsageExampleTest.

[mawiesne]

I can't review this as a human. It is impossible in this "whole bunch" fashion.

[krickert]

@mawiesne I have it in draft - so it's a WIP. If I flip it to green, it's my intent to signal that I have it ready. But if you see anything in here that concerns you LMK - I am checking the code and I'll work my hardest to hit all our standards once I flip it green.

The draft PRs are there for the gRPC progress. SInce it requires OPENNLP-1888 + other PRs it makes sense to stack on it as part of a set of features so I can test it live. This is only to demo the grpc server and test features live before writing a ton of tests - not try to rush anything.

I have zero expectation and don't want to rush anything in draft. In fact, keeping it in draft might signal there are parts not ready yet. I really like the methodical review process - so don't take my drafts as ready for review - more like "I tested it, I like what I see, I don't have major issues with it and did a few rounds of cleanup."

So I kept as draft for any feedback, concerns, and if anyone wants to collaborate on the drafts. I welcome all code suggestions and changes.

[krickert]

@mawiesne Still draft, but gettting close to "ready" - although this is fine if we put in addons. But it's based on Lucene's impl but is incredibly useful for our library.