bug: a malformed precomputed partitions file panics the index build

Description

load_precomputed_partitions in rust/lance-index/src/vector/ivf/builder.rs reads the row_id and partition columns out of a user-supplied file with .expect plus an unchecked as_primitive:

let row_ids: &UInt64Array = batch
    .column_by_name("row_id")
    .expect("malformed partition file: missing row_id column")
    .as_primitive();

A file that is missing either column, or that has the right names with the wrong physical type, panics in the middle of the build instead of returning an error. The file is the user's: LanceDataset.create_index(..., precomputed_partition_dataset=<uri>) checks only that the dataset has a single fragment and a single data file, then hands the URI to the index builder, which opens it and scans whatever schema it finds.

Nulls are a quieter version of the same problem. Both columns are read through their values buffer, which ignores validity, and the writer that produces these files marks the fields nullable. A null partition is therefore read as whatever the buffer happens to hold at that slot, so the row is assigned to an arbitrary partition and the index is built wrong with no error at all.

Steps to reproduce

import lance, pyarrow as pa

# The partitions file the user hands to create_index. row_id is Int64 here,
# which is what a naive pa.array([...]) of Python ints produces.
parts = pa.table({
    "row_id": pa.array([0, 1, 2], pa.int64()),
    "partition": pa.array([0, 1, 0], pa.uint32()),
})
lance.write_dataset(parts, "parts.lance")

ds = lance.dataset("data.lance")
ds.create_index("vector", "IVF_PQ", num_partitions=2,
                precomputed_partition_dataset="parts.lance")
# panics: malformed partition file: missing row_id column

Expected behavior

An invalid_input error naming the column, the type expected and what the file has instead. Same for a file that carries nulls in either column.

Lance version

13.0.0-beta.4 (main)

Language binding

Python, Rust

Start in rust/lance-index/src/vector/ivf/builder.rs at load_precomputed_partitions, then reproduce the issue through LanceDataset.create_index(..., precomputed_partition_dataset=...) using the Python example. Trace schema and null handling for row_id and partition. Done means malformed or nullable columns return invalid_input errors naming the problem instead of panicking or silently building an incorrect index.