feat: Implement architectural improvements (Items 7-10)

[mingjerli]

Summary

This PR implements the architectural improvements outlined in the design plan (Items 7-10):

Item 7: Path Validation

• New path_validation.py module with PathValidator class
• TOCTOU-safe file reading via _safe_read_sql_file()
• Symlink protection with opt-in allow_symlinks parameter
• Windows reserved name detection (CON, PRN, NUL, etc.)
• Unicode normalization for homoglyph attack prevention
• 100% test coverage (71 tests)

Item 10: Prompt Injection Mitigation

• New prompt_sanitization.py module with 4-layer defense
• Input sanitization with tag escaping (not removal)
• Unicode NFKC normalization for Cyrillic bypass prevention
• Output validation with semantic relevance checking
• sqlglot-based SQL validation for destructive operations
• Environment variable CLGRAPH_DISABLE_PROMPT_SANITIZATION for debugging
• 95% test coverage (100 tests)

Item 9: File Splitting

• Extract lineage_utils.py from lineage_builder.py (~592 lines)
• Extract sql_column_tracer.py from lineage_builder.py (~296 lines)
• Extract tvf_registry.py from query_parser.py (~72 lines)
• Maintain backward compatibility via re-exports

Item 8: Pipeline Decomposition

• Extract LineageTracer component (~400 lines)
• Extract MetadataManager component (~185 lines)
• Extract PipelineValidator component (~169 lines)
• Extract SubpipelineBuilder component (~183 lines)
• Pipeline now uses facade pattern with lazy initialization
• All 1,052 existing tests pass without modification

File Size Reductions

File	Before	After
pipeline.py	2,795	2,426
lineage_builder.py	3,419	2,666
query_parser.py	2,354	2,313

Test Plan

• All 1,052 existing tests pass
• 71 new path validation tests (100% coverage)
• 100 new prompt sanitization tests (95% coverage)
• 35 new module extraction tests
• 68 new component tests
• ruff check passes
• ruff format passes

🤖 Generated with Claude Code