[full-text] Use paimon-full-text in Java and Python

[JingsongLi]

Summary

Switch Paimon's Tantivy full-text global index implementation to the standalone paimon-full-text dependency for Java and PyPaimon. Single-column structured full-text DSL is delegated to the native reader, while Paimon keeps Java/Python orchestration for multi-column and hybrid query composition.

Changes

• Add the paimon-full-text-index Maven dependency and route Java full-text index reader/writer calls through org.apache.paimon.index.fulltext.
• Replace PyPaimon's tantivy-py reader/writer path with paimon_ftindex adapters, including native roaring-filter support for include_row_ids.
• Push whole single-column full-text queries down to the native reader; keep multi-match/cross-column planning in Paimon.
• Remove index-module archive layout and searcher-pool helpers that were specific to paimon-tantivy-jni.
• Update Java/Python tests and mixed-language E2E coverage for the new dependency.
• Update CI workflows to checkout, build, and install apache/paimon-full-text for Java and Python jobs.

Testing

• python -m py_compile paimon-python/pypaimon/globalindex/tantivy/tantivy_full_text_global_index_reader.py paimon-python/pypaimon/globalindex/tantivy/tantivy_full_text_index_writer.py paimon-python/pypaimon/tests/vector_search_filter_test.py paimon-python/pypaimon/tests/global_index_build_test.py paimon-python/pypaimon/tests/e2e/java_py_read_write_test.py
• python -m pytest paimon-python/pypaimon/tests/vector_search_filter_test.py paimon-python/pypaimon/tests/global_index_build_test.py -q
• PAIMON_FTINDEX_JNI_LIB_PATH=... mvn -B -ntp -pl paimon-tantivy/paimon-tantivy-index -am -Dtest=TantivyFullTextGlobalIndexTest,JavaPyTantivyE2ETest,TantivyFullTextGlobalIndexerFactoryTest -Drun.e2e.tests=true -DfailIfNoTests=false -Dcheckstyle.skip=true -Dspotless.check.skip=true clean test
• PAIMON_FTINDEX_LIB_PATH=... PYTHONPATH=... python -m pytest java_py_read_write_test.py::JavaPyReadWriteTest::test_read_tantivy_full_text_index -q
• git diff --check --cached

Notes

• Depends on [core] Support Paimon full-text query DSL paimon-full-text#6 for Paimon query DSL aliases and logical-column compatibility in the native reader.
• Python 3.6 keeps skipping the paimon_ftindex package installation because the package requires Python 3.8+.

[leaves12138]

Thanks for the cleanup and the migration to paimon-full-text. I found one Java regression around whole-query pushdown for nested multi_match; please fix it before merge. I also ran git diff --check and the focused PyPaimon full-text tests (vector_search_filter_test.py, global_index_build_test.py), which passed with 90 tests.

[leaves12138]

canPushDownWholeQuery only rejects a root MultiMatch. A single-column BooleanQuery or Boost that contains a nested MultiMatch still has columns().size() == 1, so this path pushes the whole query to TantivyFullTextGlobalIndexReader.toNativeQueryJson. The new native converter has no MultiMatch branch and throws Unsupported single-column full-text query. Before this PR, the recursive evaluator handled MultiMatch by expanding it into per-column Match queries. Please either reject nested MultiMatch in the pushdown check or add native conversion coverage, ideally with a regression test.

[leaves12138]

Thanks for addressing the nested MultiMatch pushdown regression. The recursive guard plus regression test look good to me.

Validated locally:

• git diff --check origin/master...HEAD
• PyPaimon focused full-text tests: vector_search_filter_test.py, global_index_build_test.py (90 passed)
• mvn -B -ntp -pl paimon-core -am -Pfast-build -DfailIfNoTests=false -Dtest=FullTextSearchBuilderTest#testNestedSingleColumnMultiMatchFallsBackToRecursiveEvaluation test