fix(index): align row_range with partition_size for absent partitions

[LuciferYang]

Problem

IvfModel::row_range indexed offsets and lengths directly, so a partition the model does not carry panicked. partition_size answers the same question for the same input and returns 0.

Fixes #9012.

What this changes

row_range returns an empty range where partition_size returns 0.

Worth stating plainly: I could not construct the panic through the public API. User-supplied partition ids are range-checked at load_partition, search-path ids come from find_partitions over the centroids, and neither the builder nor the distributed merger can emit a model whose num_partitions() exceeds its offsets. So this is a contract alignment between two neighbouring methods rather than a fix for a reachable failure, and it is fine to close as not worth it on those grounds.

Test plan

test_row_range_out_of_range_is_empty builds a one-partition model and checks that partition 1 and usize::MAX give an empty range and that partition_size agrees, then adds a zero-length partition to show a present-but-empty partition keeps its own offset and is still distinguishable from an absent one. Restoring the direct indexing makes it panic.

• cargo test -p lance-index --lib 1231 passed, 3 ignored
• cargo clippy --all --tests --benches -- -D warnings clean
• cargo fmt --all --check clean

[lance-gatekeeper]

✅ Gate recommendation: approve.

The prior finding is addressed: absent layout entries now produce a contextual error through the fallible range lookup, every production caller propagates it, and explicitly stored zero-length partitions retain their own offsets.