Support primary-key full-text results through scan and read

[JingsongLi]

Purpose

PyPaimon already searches primary-key full-text indexes and reads the resulting physical file positions. Native full-text scan/read and its Python binding currently expose only global row IDs, so this existing operation stays on the Python path.

Changes

• Support PK full-text queries through FullTextScan -> FullTextRead -> SearchResult, following Java's snapshot-scoped scan/read and PrimaryKeyScoredResult design.
• Reuse the existing vector/full-text shared result: global row IDs for DE, file positions for PK, original scores, serialized selected splits and projected materialization after search.
• Keep the PK FAST-only restriction, reject non-partition row filters and global row-ID allowlists, and validate plan context even for empty plans.
• Reuse the shared materializer from the existing direct PK full-text read, removing its duplicate materialization logic.
• Move shared result code into search_result modules in both core and Python bindings. Full-text bindings now return the existing SearchResult class; remove the separate global-only FullTextSearchResult wrapper and update the stub.

execute_scored() now returns the common table::SearchResult. The ranges-only execute() continues to reject PK positions, which cannot be interpreted as global row IDs. No dependency is added.

Verification

• Core tests passed: full-text 116, shared results 21, vector builders 28, hybrid search 20 (some groups overlap).
• Python binding tests: 8 passed; extension rebuilt locally.
• PyPaimon Native regression: 422 cases passed across full-text, PK full-text, vector and hybrid search; original Python tests: 59 passed.
• Java golden archives verify historical snapshots, original BM25 scores and physical positions. Tests also cover delayed/projected reads, zero projections, plan reuse, context mismatch, FAST-only errors and no retry after Native failures.
• Removing the data Parquet files temporarily still permits searching their indexes; materialization alone fails on the missing files.
• cargo fmt --all --check
• cargo +1.98.0 clippy --locked --all-targets --workspace --features fulltext,vortex -- -D warnings
• Default-feature cargo check -p paimon --locked

Companion PyPaimon PR: apache/paimon#10586. It enables this route once the Rust change reaches main.

[leaves12138]

Reviewed the scan/read split and compared the primary-key full-text path with the Java implementation, including snapshot pinning, source-file reconciliation, deletion vectors, physical-position materialization, score ordering, and the unified SearchResult API. Focused Rust full-text and primary-key tests pass, and all PR CI checks are green. No blocking issues found.