Add --json output for dry-run / preview to enable scripting and inspection

[chigichan24]

Why

• Today the dry-run path emits human-readable text only (via console.error in scripts/cli.ts:177-186).
• That makes it impossible to pipe crune's candidate list into other tools — e.g. a custom selector, a CI gate, a gh workflow, or an editor plugin.
• Once --preview (separate issue) exists, the same need applies to the synthesized output: users will want to programmatically inspect candidate Skills before installing.

What

• A new --json flag (or --format json). Initial scope: dry-run and preview modes.
• Stable, documented schema. Suggested fields per candidate:
  • topicId, label, keywords, sessionCount, sessionIds
  • reusabilityScore (number) and reusabilityScoreBreakdown (per-signal map; ties into the score-breakdown issue)
  • in preview mode: synthesizedMarkdown
• Wrap the array under a top-level object that includes a schemaVersion field for future-proofing.

How

• Centralize candidate serialization in one helper; text and JSON renderers should share the same source of truth.
• When --json is set, write JSON to stdout; keep all progress logs on stderr (matches today's separation in cli.ts, where info logs already go to console.error).
• Document the schema in docs/ and link from --help.

Notes

• Pairs naturally with the --preview and score-breakdown issues. They can be implemented in any order, but if all three land, --json --preview becomes the obvious automation entry point.