[JingsongLi]

Purpose

Indexed vector search can lose matching rows when a partition column's position in the table schema differs from its position in the partition row. For example, filtering pt in a schema id, embedding, pt can return no results when an exact scalar index filter selects a source plan.

Brief change log

Remap the partition predicate from partition-row positions back to complete table-schema positions before collecting allowed source row IDs.

Add regression coverage for single and batch vector searches in fast, full and detail modes, checking both scores and materialized rows.

Tests

• The 6 new regression cases fail against main 96e06d764365b59c9895a45692819efccf209b23 and pass with this change.
• cargo test -p paimon --lib table::de_vector_read::tests: 56 passed.
• Python binding test_vector_search.py, using a wheel built from this commit: 15 passed.
• cargo fmt --all --check and git diff --check: passed.
• Combined with [Python] Fix execution boundary regressions and restore Native CI paimon#10782, the full PyPaimon Native suite passed 23,016 tests with 89 skips.

The branch includes the dependency security updates and regenerated inventories from #1212 so CI can validate this vector fix together with its prerequisites. The functional vector patch remains byte-for-byte identical to commit 188a754573f90d113a560345edea126e999cb5f1. Merge #1212 first.

After combining the changes, the CI-pinned Rust 1.98.0 workspace Clippy check passed after rebuilding paimon, dependency-report verification passed, and all 56 vector unit tests passed again. The original vector-only CI run passed all configured build, unit and integration jobs; the combined branch has triggered a new CI run.

API and Format

No public API or storage-format changes.

Documentation

Bug fix; no documentation changes required.

[leaves12138]

LGTM