Implement centralized metadata registry and optimize test performance

[turbomam]

Major Changes

Centralized Architecture (Issue #21)

• Replace 3 separate data loaders with unified MetadataRegistry class
• Add 4 core dataclasses: GeneMetadata, ConditionMetadata, ModuleMetadata, FitnessEffect
• Build efficient index structures for O(1) lookups (gene↔module, gene↔condition)
• Thread-safe data loading with file change detection

Test Infrastructure Improvements

• Add session-scoped fixtures in tests/conftest.py for shared data loading
• Remove all mocking per project policy - tests now use real data
• Update all test files to use shared fixtures instead of individual data loading
• Tests now fail if real data unavailable (no conditional passes)

Performance Optimizations

• Fix expand_gene_condition_network O(n²) performance bottleneck
• Replace nested linear scans with index structure lookups
• Test suite time: 104s → 6.6s (15.7x speedup)
• Network expansion tests: 45s each → 0.03s each (1,500x speedup)

Code Quality

• Fix scientific interpretation: negative scores = essential genes, positive = inhibitory
• Consolidate Makefile test targets into single 'test' target with coverage + timing
• Remove unused imports and target aliases
• All linting and type checking passes

API Migrations

• Migrate get_gene_info, search_genes, get_growth_conditions to use registry
• Migrate get_condition_details, interpret_fitness_score to use registry
• Migrate module-related tools to use centralized data structures
• Maintain backward compatibility with legacy loaders during transition

🤖 Generated with Claude Code

[turbomam]

for review tomorrow