Implement incremental validation — only re-analyze changed files

[rudra496]

Summary

The cache system exists (cache.py) but currently only caches file hashes. Implement true incremental validation that:

1. Tracks file modification times and content hashes
2. Only re-analyzes files that changed since last validation
3. Supports dependency-aware invalidation (if a.py imports b.py and b.py changes, re-analyze a.py too)

Current State

The cache.py module exists but needs enhancement. The multi_file.py handles dependency analysis.

Implementation Plan

1. Content-addressed cache: Store {file_hash: ValidationResult} instead of {file_path: ValidationResult}
2. Dependency graph: Build import graph, invalidate dependents when a file changes
3. Incremental mode flag: aitrust validate src/ --incremental
4. Cache manifest: Store .aitrust_cache/manifest.json with file paths, hashes, and timestamps
5. Smart invalidation: Only re-analyze if file content hash changed

Performance Target

• First run: same as current (full analysis)
• Second run (no changes): <100ms (just cache lookup)
• Second run (1 file changed): analyze only that file + dependents

Acceptance Criteria

• --incremental flag works
• Unchanged files served from cache
• Dependency-aware invalidation
• Cache manifest persisted between runs
• Benchmark showing >10x speedup on unchanged codebases
• Tests

Difficulty

Advanced — requires careful cache invalidation logic and dependency tracking.