Support batch vector query API and shared flat KNN scan

[BubbleCal]

Parent: #6820
Linear: https://linear.app/lancedb/issue/OSS-923/support-batch-vector-query-api-and-shared-flat-knn-scan
Linear parent: https://linear.app/lancedb/issue/OSS-922/vector-search-performance-optimization-roadmap
Related historical issue: #876

Goal

Allow one vector search request to contain multiple query vectors and make the first implementation immediately useful for flat KNN by sharing scan/load work across the query batch.

Scope

• Add or finalize the public API shape for batch query vectors in Rust and Python.
• Return results with an explicit per-query discriminator, for example query_index, so callers can split top-k results by query vector.
• Apply limit/top-k per query, not globally across all query vectors.
• For flat KNN, load each data batch once and evaluate all query vectors against that loaded batch.
• It is acceptable for the first version to keep ANN/indexed search out of scope, as long as the behavior is explicit and covered by tests.

Acceptance criteria

• Rust tests compare batch query results against repeated single-query results.
• Python tests cover 2-D query input and validate output grouping by query id.
• Docs/docstrings explain the batch input shape and result contract.
• A small flat KNN benchmark demonstrates lower tail latency versus issuing repeated independent queries, especially when I/O or decode cost dominates.

[LeoReeYang]

hey, I would like to take this one