Add Duplicate Clustering Output Mode

[tim-dickey]

Goal

Group duplicates into clusters (connected components) rather than just listing pairwise matches to simplify remediation workflows.

Scope

• Build undirected graph where edges connect files with similarity >= threshold.
• Derive clusters via DFS/Union-Find.
• CLI flag: --clusters to output cluster report.
• JSON output structure: list of clusters with member files + representative (largest or earliest path).
• Table output: cluster id, size, representative, member count.

Acceptance Criteria

• Cluster mode outputs deterministic ordering (by representative path).
• Tests: cluster formation on synthetic dataset (multiple interconnected duplicates).
• Documentation: README + schema update with cluster format.

Non-Goals

• Advanced representative selection (simple heuristic only).

Future

• Cluster similarity summary stats (avg/max/min within cluster).

[tim-dickey]

Implemented MinHash + LSH prefilter:

• Added minhash.py with minhash_signature and lsh_candidates.
• Integrated prefilter into DuplicateFinder.find_duplicates (auto-skips for <50 files).
• CLI flags: --prefilter --minhash-perms --lsh-bands.
• Tests: test_minhash_prefilter.py covers equivalence + skip small datasets.
• README updated with usage guidance and trade-offs.
  Next enhancements:
• Parameter tuning heuristics.
• False-negative probability estimation doc section.
• Adaptive band sizing based on shingle set cardinality.