feat!: support sampling selected fragments

[Xuanwo]

This extends Dataset::sample to accept optional fragment IDs so grouped training jobs can sample from selected fragments without reimplementing deletion-aware row offset mapping. It rejects empty fragment selections and unknown fragment IDs while preserving the existing behavior for whole-dataset sampling.

[github-actions]

PR Review

Clean, well-structured change. The refactoring of row_offsets_to_row_addresses to accept &[FileFragment] instead of &Dataset is a nice generalization that enables the subset sampling without duplicating logic.

One minor observation (not blocking)

Double row counting for fragment-subset path: When fragment_ids is Some, count_rows is called for each selected fragment twice — once in the buffer_unordered(16) fold to compute num_rows, and again sequentially inside row_offsets_to_row_addresses. If count_rows isn't cached, this doubles the I/O for deletion vector reads. You could pass the precomputed total into row_offsets_to_row_addresses or pre-collect the per-fragment counts, but it's likely negligible for typical workloads.

Tests

Good coverage: subset sampling with deletions, empty fragment list rejection, unknown fragment ID rejection. The duplicate fragment ID in test_sample_with_fragment_ids ([0, 0, 2]) is a nice touch for exercising the dedup via BTreeSet.

LGTM.

[codecov]

Codecov Report

❌ Patch coverage is 87.93103% with 7 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
rust/lance/src/dataset.rs	90.38%	0 Missing and 5 partials ⚠️
rust/lance/src/dataset/take.rs	66.66%	0 Missing and 1 partial ⚠️
rust/lance/src/index/vector/utils.rs	0.00%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[wjones127]

This looks good. Have a few minor suggestions on tests, but those are optional.