Add PI workflow skills for task-aware orchestration

[GTC2333]

Summary

• add a PI decision-framing Skill that maps outcome-sensitive open choices before delegation
• add distinct PI workflow Skills for model selection, data analysis, literature synthesis, study design, and software delivery
• expose the Skills through the bundled research plugin, including a short instruction pointer for custom PI personas
• make PI load every applicable workflow Skill for mixed tasks instead of treating one router match as the workflow
• update role-targeted skill-path and plugin tests

Why

A Sleep-EDF trajectory treated a detailed single-model deliverable as a bounded implementation task. It skipped Librarian/Experimentalist, used one Engineer feasibility run as completion evidence, and never compared alternatives. The new decision frame explicitly separates delivery constraints and design hints from scientific choices that remain open.

Verification

• six Skills pass quick_validate.py
• independent forward tests: open Sleep model design, fixed EEGNet, literature-only research, and CSV parser debugging route to distinct workflows
• npx tsc -b packages/plugin-sdk packages/protocol packages/skills packages/runtime packages/backend-core packages/cli
• Runtime: 703 tests passed
• CLI: 108 tests passed
• npm pack --dry-run -w @brainpilot/plugin-research includes all new prompts and Skills