v0.3.0 — models, compare, sharpen

Adds the CLI surface area required by the sc-kg-analysis skills (sc-models, sc-compare, sc-reframe, sc-unstuck, sc-setup) that previously referenced commands the CLI didn't expose.

New commands

• mentor models [--search <term>] — list / filter mental models
• mentor compare <options...> [--context <context>] — compare options, surfacing decision context first
• mentor sharpen --situation <text> --current-thinking <text> [--what-feels-risky <text>] — pressure-test existing thinking

Improvements

• mentor model "<name>" — resolve by label, not just mm-N or a bare number
• mentor search --top <n> — pass top_k through to /retrieve (1–20)
• mentor --version now reads from package.json instead of a hardcoded string

compare and sharpen are pure compositions of /retrieve calls that mirror the server-side skill handlers, so no new API endpoints are required.