[Forge] Seed neurodegeneration ML benchmark registry — 5 prediction tasks with answer keys open

**Problem:** SciDEX has 0 benchmarks in its artifact registry (confirmed via `SELECT COUNT(*) FROM artifacts WHERE artifact_type = 'benchmark'`). Without benchmarks, the platform cannot demonstrate predictive validity, cannot attract computational biology collaborators, and cannot self-evaluate whether its hypotheses have real predictive power. **Goal:** Create 5 neurodegeneration ML benchmarks as registered artifacts, each with a task definition, dataset pointer, baseline performance, and answer key format. **Benchmarks to create:** 1. **AD Protein Aggregation Propensity** — predict which proteins are prone to pathological aggregation in Alzheimer's disease. Positive set: known Aβ/tau interactors; negative set: non-aggregating brain proteins. Task: binary classification. 2. **PD Dopaminergic Neuron Expression Signature** — predict which genes are differentially expressed in dopaminergic neurons in PD vs controls. Use GTEx brain expression as baseline. 3. **ALS Motor Neuron Survival Gene Prediction** — identify genes whose knockout protects or sensitizes motor neurons in ALS models. Task: rank genes by survival effect direction. 4. **Neurodegeneration Drug Target Druggability** — predict druggability (structural + ligandability) for the top 100 neurodegeneration KG entities. Task: regression score 0-1. 5. **Glymphatic Clearance Biomarker Panel** — predict which CSF/blood biomarkers best discriminate Alzheimer's progression stages. Task: feature importance ranking. **Implementation for each benchmark:** 1. Define benchmark spec as JSON (task, metric, dataset_source, baseline_accuracy, answer_key_format) 2. Register in artifacts table: `INSERT INTO artifacts (title, artifact_type, description, content_json, provenance) VALUES (