Add Rust optimization examples from squeeze library (HNSW + t-SNE)

[GeorgePearse]

Summary

This PR adds two new Genesis optimization examples targeting algorithms from the squeeze high-performance Rust library for dimensionality reduction:

• HNSW Search Optimization: Targets the beam search algorithm in HNSW graphs for approximate nearest neighbor search
• t-SNE Gradient Optimization: Targets the O(n²) gradient computation that dominates t-SNE embedding time
• ROADMAP.md: Comprehensive language support documentation with execution backend matrix

Results

HNSW (100 generations, in progress)

Metric	Baseline	Best Found	Improvement
Score	53.79	61.32	+14%
Key optimization	HashSet visited	Vec visited	-

t-SNE (50 generations, in progress)

Metric	Baseline	Best Found	Improvement
Score	149.50	155.08	+3.7%
Key optimization	Standard loops	Blocked unroll + symmetry	-

Files Changed

New Examples

• examples/squeeze_hnsw/initial.rs - Standalone HNSW implementation with EVOLVE-BLOCK markers
• examples/squeeze_hnsw/evaluate.py - Fitness evaluation (recall@10 × speed)
• examples/squeeze_tsne/initial.rs - Standalone t-SNE gradient implementation
• examples/squeeze_tsne/evaluate.py - Fitness evaluation (trustworthiness × speed)

New Configs

• configs/task/squeeze_hnsw.yaml - Task definition with optimization hints
• configs/task/squeeze_tsne.yaml - Task definition with optimization hints
• configs/variant/squeeze_hnsw_example.yaml - Local execution variant
• configs/variant/squeeze_hnsw_e2b.yaml - E2B cloud execution variant
• configs/variant/squeeze_tsne_example.yaml - Local execution variant

Documentation

• ROADMAP.md - Language support matrix, execution backends, planned features
• README.md - Added roadmap link to docs table

Test plan

• HNSW evaluator runs locally
• t-SNE evaluator runs locally
• Genesis experiments launch successfully
• Experiments show improvement over baseline

🤖 Generated with Claude Code