fix: shift offsets after trimming values in merge_with_schema

[kaan-simbe]

Summary

• Fix panic InvalidArgumentError("Max offset of X exceeds length of values Y") in ListArray::new during to_table(filter=..., columns=[list_struct_col, ...]) on v2.0 datasets.
• Root cause: merge_with_schema (called from TakeStream::map_batch) passed left_list.trimmed_values() alongside left_list.offsets().clone(). When the left list is a sliced view (e.g. a filtered batch), offsets do not start at zero and reference positions past the end of the trimmed child, panicking in ListArray::new.
• Add ListArrayExt::trimmed_offsets() that returns offsets shifted to start at zero, and use it in the List/LargeList branches of merge_with_schema.

Test plan

• New regression test test_merge_with_schema_sliced_list_struct in lance-arrow: fails on main with the exact panic, passes with the fix.
• All existing lance-arrow merge tests still pass (9/9).
• Python repro from the issue (1M rows, 200k + 800k + 14650 sparse-tail pattern) no longer panics and returns the expected 214,650 rows with correct data (verified against a manually-filtered reference batch).

Fixes #6580

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[Xuanwo]

Thanks, I agree with the root cause and the fix direction.

Could you also add a higher-level regression through the scan/take path? The new unit test covers the local invariant, but the user-facing failure is to_table(filter=..., columns=[list_struct_col]), which goes through TakeExec -> merge_with_schema. A LargeList case would also be useful since this PR changes that branch too.

[kaan-simbe]

Thanks for the review @Xuanwo!

Added in aecf0bc:

• Refactored the lance-arrow test into a generic helper + thin i32/i64 wrappers (matching the existing test_merge_struct_lists_generic pattern), so the LargeList branch is covered too.
• Added a higher-level regression in scanner.rs that goes through the full scan().filter().project([list_struct_col]) path. To get the planner to actually emit TakeExec with items split across input and take (which is what triggers the List<Struct> merge branch), I had to nudge it with MaterializationStyle::AllEarlyExcept([items.b]). Parametrized over V2_0 / Stable / V2_2 since the panic is v2-specific.

One note: I couldn't shrink the dataset below ~500k rows and still reproduce the panic — at ~250k it stops triggering, so the regression test runs ~0.7s for all 3 cases.

[kaan-simbe]

@Xuanwo anything else I can do here?

[xloya]

We encountered a similar problem in our production environment, and the issue was resolved after trying to cherry pick this PR.