fix(table): guard DV filtering against stale row counts

[fallintoplace]

Summary

Prevent deletion-vector filtering from panicking when a data file emits more rows than its manifest record_count.

Problem

The contiguous fast path builds a keep mask sized from DataFile.Count() and slices it using the rows emitted by each Arrow batch. If the manifest count is stale or otherwise smaller than the physical file, the batch end falls outside the mask and array.NewSlice panics.

The row-group-pruned path already handles this mismatch by preserving positions beyond the mask. The contiguous path should behave the same way.

Fix

Keep the existing zero-copy mask slice while a batch is within the declared row count. If a batch crosses that boundary, build a bounded mask for that batch and keep rows whose positions are beyond the declared count. Once a batch is entirely beyond the mask, pass it through without allocating an all-true filter. In-range deletion-vector positions are still applied normally.

Tests

Added coverage across three batches: one within the manifest count, one crossing it, and one entirely beyond it. The test verifies in-range deletes while preserving rows beyond the declared count.

go test ./table -count=1