refactor: Extract sort-merge join filter logic into separate module

Refactored the sort-merge join implementation to improve code organization
by extracting all filter-related logic into a dedicated filter.rs module.

Changes:
- Created new filter.rs module (~576 lines) containing:
  - Filter metadata tracking (FilterMetadata struct)
  - Deferred filtering decision logic (needs_deferred_filtering)
  - Filter mask correction for different join types (get_corrected_filter_mask)
  - Filter application with null-joined row handling (filter_record_batch_by_join_type)
  - Helper functions for filter column extraction and batch filtering

- Updated stream.rs:
  - Removed ~450 lines of filter-specific code
  - Now delegates to filter module functions
  - Simplified main join logic to focus on stream processing

- Updated tests.rs:
  - Updated imports to use new filter module
  - Changed test code to use FilterMetadata struct
  - All 47 sort-merge join tests passing

- Fixed null-joined batch creation for joins with different column counts:
  - Correctly handles LEFT/RIGHT/FULL outer joins with asymmetric schemas
  - Properly extracts and replaces columns based on join type and batch organization
  - Uses RecordBatchOptions to handle nullable field validation in outer joins

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Author: viirya