[codex] Add AlphaGenome embedding and PCA workflow

[neonine2]

Summary

Adds the missing Deepdraw workflow steps for embedding generation and PCA reduction, fully contained in this repository.

• Adds deepdraw embed-alphagenome for AlphaGenome embedding extraction from a design-pool CSV or FASTA.
• Adds deepdraw pca with kneedle/variance/elbow/L-method component selection.
• Adds an isolated envs/alphagenome uv environment for AlphaGenome/JAX/Hugging Face dependencies.
• Updates the README end-to-end workflow, including auth, embedding, PCA, init, suggest, and recommended embedding hardware.
• Adds focused tests for embedding loading, PCA reduction, AlphaGenome input parsing, and CLI delegation.

Verification

• Real AlphaGenome inference on a two-sequence smoke pool using the downloaded gated checkpoint produced raw embeddings shaped (2, 3072).
• Real deepdraw pca on those embeddings produced (2, 2) reduced embeddings.
• Real deepdraw init consumed the PCA output and wrote round_000_to_measure.csv.
• Real deepdraw suggest consumed a measurement CSV and wrote round_001_to_measure.csv.
• uv run pytest passed: 191 passed, 7 warnings.
• Pre-commit hooks passed during commit.

Notes

Apple Silicon is supported through JAX CPU for this AlphaGenome path. JAX Metal could see the M5 Max GPU locally, but the current Metal backend failed on basic allocation/StableHLO paths, so the README recommends Linux + CUDA GPU for production embedding throughput.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!