Make convergence clustering threshold configurable

[that-github-user]

Problem

The similarity threshold used to cluster agents into convergence groups is hard-coded at 0.3 in src/scoring/convergence.ts:

const SIMILARITY_THRESHOLD = 0.3;

This value was chosen without project-specific justification. For tasks with many small, similar solutions (e.g. adding a type annotation) a threshold of 0.3 clusters too aggressively. For tasks with large divergent rewrites, 0.3 may be too lenient. There is no way for users to tune this without editing source code.

Proposed solution

• Add --convergence-threshold <0.0–1.0> flag to thinktank run (default: 0.3, preserving current behavior)
• Thread it through RunOptions → analyzeConvergence()
• Document what the value means (higher = stricter grouping = more, smaller clusters)
• Add a brief note in README or help text: --convergence-threshold 0.5 is good for refactoring tasks; 0.2 works for exploratory tasks.

Acceptance criteria

• --convergence-threshold 0.7 changes clustering behavior visibly in output
• Value validated: must be between 0.0 and 1.0 inclusive
• Default is 0.3 (no change in behavior without the flag)
• RunOptions type updated
• Convergence tests updated to pass threshold explicitly (no test should rely on the hard-coded default)

[that-github-user]

Fixed in #100. Convergence threshold now configurable via --threshold flag (default 0.3). Generated by thinktank Opus (5 agents, 76% consensus).