Establish the typed BrainlessLab research platform

[btgaskin]

Summary

• establish the typed composition and evaluation kernel around NodeSpec, TaskSpec, CompositionSpec, EvaluationSpec, and named evaluation targets
• add the five standard research operations: profile, sweep, ablation, evolution, and benchmark
• add versioned experiment definitions and portable record bundles with trial tables, seed ledgers, summaries, figures, and HTML reports
• keep Tracking and Pong as the initial core benchmark; register the four Plank CartPole levels as experimental challenge tasks
• consolidate the public guide, examples, skills, archived research material, CI, package metadata, and interactive task demos around the new contracts

Why

This gives BrainlessLab one validated path from a registered node and task, through a runnable composition and evaluation protocol, to a reproducible operation record. It removes parallel plan and documentation seams before broader experimental work begins.

Population-coded sensorimotor interfaces and open CartPole design evolution are intentionally excluded. They will be developed from this merged base in a separate PR.

Validation

• julia --project=. -e 'using Pkg; Pkg.test()' — passed
• cd site && bun run build — passed; 39 pages built
• git diff --check origin/main...HEAD — passed