Split SplitBy::Layout into Layout and LayoutSubSplitting variants

[robert3005]

Rationale for this change

This change separates the concerns of chunk-boundary-aware splitting from automatic sub-division of wide chunk spans. Previously, SplitBy::Layout combined both behaviors: it would split at chunk boundaries and also sub-divide spans wider than MAX_SPLIT_ROWS to enable intra-file decode parallelism.

The new design introduces two distinct variants:

• SplitBy::Layout: Splits only at chunk boundaries, yielding fewer, larger batches that follow the file's chunking exactly
• SplitBy::LayoutSubSplitting: The previous default behavior—splits at chunk boundaries and sub-divides wide spans for parallelism

This makes the splitting strategy more explicit and allows users to choose between fine-grained parallelism (default) or coarser-grained batching that respects the file's natural structure.

What changes are included in this PR?

Core changes:

• Split SplitBy::Layout into two variants in vortex-layout/src/scan/split_by.rs
• Updated SplitBy::Layout to only split at chunk boundaries without sub-division
• Made SplitBy::LayoutSubSplitting the new default (with #[default])
• Modified the split logic to conditionally apply subdivide_large_spans() only for LayoutSubSplitting

API additions:

• Added ScanBuilder::with_no_sub_splitting() convenience method in vortex-layout/src/scan/scan_builder.rs to explicitly opt into SplitBy::Layout

Test improvements:

• Refactored test infrastructure: extracted a reusable StubReader struct to replace the inline DupReader
• Added test_layout_keeps_large_chunk_whole() to verify that SplitBy::Layout keeps wide chunks intact while SplitBy::LayoutSubSplitting sub-divides them
• Updated existing tests to use the new variants appropriately

Integration updates:

• Updated vortex-file/src/file.rs to use LayoutSubSplitting for the default splits() method
• Updated vortex-datafusion/src/persistent/opener.rs to use LayoutSubSplitting for natural split ranges
• Updated vortex-python/src/dataset.rs to default to LayoutSubSplitting
• Updated documentation comments to reflect the new variant names

What APIs are changed? Are there any user-facing changes?

Public API changes:

• SplitBy enum now has two layout-related variants instead of one
• ScanBuilder::with_no_sub_splitting() is a new convenience method for opting into SplitBy::Layout
• Default scan behavior remains unchanged (still uses sub-splitting), but is now explicitly SplitBy::LayoutSubSplitting

User-facing impact:

• Users who want to disable sub-splitting can now call .with_no_sub_splitting() on a scan builder
• The default behavior is unchanged—scans still sub-divide wide chunks for parallelism
• File splits reported by VortexFile::splits() now reflect sub-divided boundaries (matching the new default)

https://claude.ai/code/session_01M9d52UfuZRn3cC9eixruzx