Support supplied global-index results in batch scans

[JingsongLi]

Purpose

Support the existing PyPaimon with_global_index_result read path in Rust core. Primary-key vector/full-text results carry physical IndexedSplits, not global row IDs. Ordinary scans currently cannot consume those captured selections, forcing the Python scanner path.

Match Java PrimaryKeyBatchScan and DataEvolutionBatchScan: preserve PK result snapshots/splits before ordinary planning, and apply DE row-ID selections and scores to the final planned ranges. Query-authorized scans ignore external raw-index selections.

PyPaimon integration: apache/paimon#10591.

Brief change log

• Unify table vector/batch-vector/full-text/hybrid search results and scan selections as one core GlobalIndexResult, following Java. Retain optional source context internally for ranked reads.
• Add TableScan::with_global_index_result and IncrementalScan::with_global_index_result; searches pass their result directly without a conversion API.
• Keep DE incremental snapshot windows and final score alignment in core; reject incompatible coordinate spaces and selection modes.
• Retain scored Java IndexedSplit frames, validate score lengths and truncated payloads, and preserve absent totalBuckets rather than materializing it during serialization.
• Expose a single Python GlobalIndexResult and GlobalIndexResultReadBuilder, with thin static transports for external ranges/scores/splits. Remove the separate table SearchResult, Python HybridSearchResult, and explicit conversion. Update type stubs and C/DataFusion callers.

Tests

• 14 global-index-result tests covering direct handoff, ranked reads, source snapshots, external read-context rejection, score alignment, incremental windows, authorization, and invalid inputs.
• 490 vector-related, 116 full-text-related, and 20 hybrid core tests passed.
• 188 Python binding vector/read/resolved-table tests passed.
• Cross-repository PyPaimon: 132 planner/result tests plus 46 subtests, and 188 PK full-text/vector/hybrid search tests passed against the locally built binding.
• All four search families return GlobalIndexResult and support ranked reads and direct scan handoff, with exact final split-score transport checks.
• cargo +1.98.0 clippy --locked --all-targets --workspace --features fulltext,vortex -- -D warnings
• cargo fmt --all --check

API and Format

Adds the core selection APIs and unifies the public table-search result type as GlobalIndexResult (no compatibility aliases). Existing plain-split callers keep their behavior. No table storage-format changes; the split wire reader now accepts the existing Java float-score representation.

Documentation

Updated API documentation and Python type stubs.

[QuakeWang]

+1